Actual source code: aij.c
petsc-3.5.1 2014-07-24
2: /*
3: Defines the basic matrix operations for the AIJ (compressed row)
4: matrix storage format.
5: */
8: #include <../src/mat/impls/aij/seq/aij.h> /*I "petscmat.h" I*/
9: #include <petscblaslapack.h>
10: #include <petscbt.h>
11: #include <petsc-private/kernels/blocktranspose.h>
12: #if defined(PETSC_THREADCOMM_ACTIVE)
13: #include <petscthreadcomm.h>
14: #endif
18: PetscErrorCode MatGetColumnNorms_SeqAIJ(Mat A,NormType type,PetscReal *norms)
19: {
21: PetscInt i,m,n;
22: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)A->data;
25: MatGetSize(A,&m,&n);
26: PetscMemzero(norms,n*sizeof(PetscReal));
27: if (type == NORM_2) {
28: for (i=0; i<aij->i[m]; i++) {
29: norms[aij->j[i]] += PetscAbsScalar(aij->a[i]*aij->a[i]);
30: }
31: } else if (type == NORM_1) {
32: for (i=0; i<aij->i[m]; i++) {
33: norms[aij->j[i]] += PetscAbsScalar(aij->a[i]);
34: }
35: } else if (type == NORM_INFINITY) {
36: for (i=0; i<aij->i[m]; i++) {
37: norms[aij->j[i]] = PetscMax(PetscAbsScalar(aij->a[i]),norms[aij->j[i]]);
38: }
39: } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Unknown NormType");
41: if (type == NORM_2) {
42: for (i=0; i<n; i++) norms[i] = PetscSqrtReal(norms[i]);
43: }
44: return(0);
45: }
49: PetscErrorCode MatFindOffBlockDiagonalEntries_SeqAIJ(Mat A,IS *is)
50: {
51: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
52: PetscInt i,m=A->rmap->n,cnt = 0, bs = A->rmap->bs;
53: const PetscInt *jj = a->j,*ii = a->i;
54: PetscInt *rows;
55: PetscErrorCode ierr;
58: for (i=0; i<m; i++) {
59: if ((ii[i] != ii[i+1]) && ((jj[ii[i]] < bs*(i/bs)) || (jj[ii[i+1]-1] > bs*((i+bs)/bs)-1))) {
60: cnt++;
61: }
62: }
63: PetscMalloc1(cnt,&rows);
64: cnt = 0;
65: for (i=0; i<m; i++) {
66: if ((ii[i] != ii[i+1]) && ((jj[ii[i]] < bs*(i/bs)) || (jj[ii[i+1]-1] > bs*((i+bs)/bs)-1))) {
67: rows[cnt] = i;
68: cnt++;
69: }
70: }
71: ISCreateGeneral(PETSC_COMM_SELF,cnt,rows,PETSC_OWN_POINTER,is);
72: return(0);
73: }
77: PetscErrorCode MatFindZeroDiagonals_SeqAIJ_Private(Mat A,PetscInt *nrows,PetscInt **zrows)
78: {
79: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
80: const MatScalar *aa = a->a;
81: PetscInt i,m=A->rmap->n,cnt = 0;
82: const PetscInt *jj = a->j,*diag;
83: PetscInt *rows;
84: PetscErrorCode ierr;
87: MatMarkDiagonal_SeqAIJ(A);
88: diag = a->diag;
89: for (i=0; i<m; i++) {
90: if ((jj[diag[i]] != i) || (aa[diag[i]] == 0.0)) {
91: cnt++;
92: }
93: }
94: PetscMalloc1(cnt,&rows);
95: cnt = 0;
96: for (i=0; i<m; i++) {
97: if ((jj[diag[i]] != i) || (aa[diag[i]] == 0.0)) {
98: rows[cnt++] = i;
99: }
100: }
101: *nrows = cnt;
102: *zrows = rows;
103: return(0);
104: }
108: PetscErrorCode MatFindZeroDiagonals_SeqAIJ(Mat A,IS *zrows)
109: {
110: PetscInt nrows,*rows;
114: *zrows = NULL;
115: MatFindZeroDiagonals_SeqAIJ_Private(A,&nrows,&rows);
116: ISCreateGeneral(PetscObjectComm((PetscObject)A),nrows,rows,PETSC_OWN_POINTER,zrows);
117: return(0);
118: }
122: PetscErrorCode MatFindNonzeroRows_SeqAIJ(Mat A,IS *keptrows)
123: {
124: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
125: const MatScalar *aa;
126: PetscInt m=A->rmap->n,cnt = 0;
127: const PetscInt *ii;
128: PetscInt n,i,j,*rows;
129: PetscErrorCode ierr;
132: *keptrows = 0;
133: ii = a->i;
134: for (i=0; i<m; i++) {
135: n = ii[i+1] - ii[i];
136: if (!n) {
137: cnt++;
138: goto ok1;
139: }
140: aa = a->a + ii[i];
141: for (j=0; j<n; j++) {
142: if (aa[j] != 0.0) goto ok1;
143: }
144: cnt++;
145: ok1:;
146: }
147: if (!cnt) return(0);
148: PetscMalloc1((A->rmap->n-cnt),&rows);
149: cnt = 0;
150: for (i=0; i<m; i++) {
151: n = ii[i+1] - ii[i];
152: if (!n) continue;
153: aa = a->a + ii[i];
154: for (j=0; j<n; j++) {
155: if (aa[j] != 0.0) {
156: rows[cnt++] = i;
157: break;
158: }
159: }
160: }
161: ISCreateGeneral(PETSC_COMM_SELF,cnt,rows,PETSC_OWN_POINTER,keptrows);
162: return(0);
163: }
167: PetscErrorCode MatDiagonalSet_SeqAIJ(Mat Y,Vec D,InsertMode is)
168: {
170: Mat_SeqAIJ *aij = (Mat_SeqAIJ*) Y->data;
171: PetscInt i,*diag, m = Y->rmap->n;
172: MatScalar *aa = aij->a;
173: PetscScalar *v;
174: PetscBool missing;
177: if (Y->assembled) {
178: MatMissingDiagonal_SeqAIJ(Y,&missing,NULL);
179: if (!missing) {
180: diag = aij->diag;
181: VecGetArray(D,&v);
182: if (is == INSERT_VALUES) {
183: for (i=0; i<m; i++) {
184: aa[diag[i]] = v[i];
185: }
186: } else {
187: for (i=0; i<m; i++) {
188: aa[diag[i]] += v[i];
189: }
190: }
191: VecRestoreArray(D,&v);
192: return(0);
193: }
194: MatSeqAIJInvalidateDiagonal(Y);
195: }
196: MatDiagonalSet_Default(Y,D,is);
197: return(0);
198: }
202: PetscErrorCode MatGetRowIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *m,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done)
203: {
204: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
206: PetscInt i,ishift;
209: *m = A->rmap->n;
210: if (!ia) return(0);
211: ishift = 0;
212: if (symmetric && !A->structurally_symmetric) {
213: MatToSymmetricIJ_SeqAIJ(A->rmap->n,a->i,a->j,ishift,oshift,(PetscInt**)ia,(PetscInt**)ja);
214: } else if (oshift == 1) {
215: PetscInt *tia;
216: PetscInt nz = a->i[A->rmap->n];
217: /* malloc space and add 1 to i and j indices */
218: PetscMalloc1((A->rmap->n+1),&tia);
219: for (i=0; i<A->rmap->n+1; i++) tia[i] = a->i[i] + 1;
220: *ia = tia;
221: if (ja) {
222: PetscInt *tja;
223: PetscMalloc1((nz+1),&tja);
224: for (i=0; i<nz; i++) tja[i] = a->j[i] + 1;
225: *ja = tja;
226: }
227: } else {
228: *ia = a->i;
229: if (ja) *ja = a->j;
230: }
231: return(0);
232: }
236: PetscErrorCode MatRestoreRowIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done)
237: {
241: if (!ia) return(0);
242: if ((symmetric && !A->structurally_symmetric) || oshift == 1) {
243: PetscFree(*ia);
244: if (ja) {PetscFree(*ja);}
245: }
246: return(0);
247: }
251: PetscErrorCode MatGetColumnIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *nn,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done)
252: {
253: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
255: PetscInt i,*collengths,*cia,*cja,n = A->cmap->n,m = A->rmap->n;
256: PetscInt nz = a->i[m],row,*jj,mr,col;
259: *nn = n;
260: if (!ia) return(0);
261: if (symmetric) {
262: MatToSymmetricIJ_SeqAIJ(A->rmap->n,a->i,a->j,0,oshift,(PetscInt**)ia,(PetscInt**)ja);
263: } else {
264: PetscCalloc1(n+1,&collengths);
265: PetscMalloc1((n+1),&cia);
266: PetscMalloc1((nz+1),&cja);
267: jj = a->j;
268: for (i=0; i<nz; i++) {
269: collengths[jj[i]]++;
270: }
271: cia[0] = oshift;
272: for (i=0; i<n; i++) {
273: cia[i+1] = cia[i] + collengths[i];
274: }
275: PetscMemzero(collengths,n*sizeof(PetscInt));
276: jj = a->j;
277: for (row=0; row<m; row++) {
278: mr = a->i[row+1] - a->i[row];
279: for (i=0; i<mr; i++) {
280: col = *jj++;
282: cja[cia[col] + collengths[col]++ - oshift] = row + oshift;
283: }
284: }
285: PetscFree(collengths);
286: *ia = cia; *ja = cja;
287: }
288: return(0);
289: }
293: PetscErrorCode MatRestoreColumnIJ_SeqAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done)
294: {
298: if (!ia) return(0);
300: PetscFree(*ia);
301: PetscFree(*ja);
302: return(0);
303: }
305: /*
306: MatGetColumnIJ_SeqAIJ_Color() and MatRestoreColumnIJ_SeqAIJ_Color() are customized from
307: MatGetColumnIJ_SeqAIJ() and MatRestoreColumnIJ_SeqAIJ() by adding an output
308: spidx[], index of a->a, to be used in MatTransposeColoringCreate_SeqAIJ() and MatFDColoringCreate_SeqXAIJ()
309: */
312: PetscErrorCode MatGetColumnIJ_SeqAIJ_Color(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *nn,const PetscInt *ia[],const PetscInt *ja[],PetscInt *spidx[],PetscBool *done)
313: {
314: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
316: PetscInt i,*collengths,*cia,*cja,n = A->cmap->n,m = A->rmap->n;
317: PetscInt nz = a->i[m],row,*jj,mr,col;
318: PetscInt *cspidx;
321: *nn = n;
322: if (!ia) return(0);
324: PetscCalloc1(n+1,&collengths);
325: PetscMalloc1((n+1),&cia);
326: PetscMalloc1((nz+1),&cja);
327: PetscMalloc1((nz+1),&cspidx);
328: jj = a->j;
329: for (i=0; i<nz; i++) {
330: collengths[jj[i]]++;
331: }
332: cia[0] = oshift;
333: for (i=0; i<n; i++) {
334: cia[i+1] = cia[i] + collengths[i];
335: }
336: PetscMemzero(collengths,n*sizeof(PetscInt));
337: jj = a->j;
338: for (row=0; row<m; row++) {
339: mr = a->i[row+1] - a->i[row];
340: for (i=0; i<mr; i++) {
341: col = *jj++;
342: cspidx[cia[col] + collengths[col] - oshift] = a->i[row] + i; /* index of a->j */
343: cja[cia[col] + collengths[col]++ - oshift] = row + oshift;
344: }
345: }
346: PetscFree(collengths);
347: *ia = cia; *ja = cja;
348: *spidx = cspidx;
349: return(0);
350: }
354: PetscErrorCode MatRestoreColumnIJ_SeqAIJ_Color(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscInt *spidx[],PetscBool *done)
355: {
359: MatRestoreColumnIJ_SeqAIJ(A,oshift,symmetric,inodecompressed,n,ia,ja,done);
360: PetscFree(*spidx);
361: return(0);
362: }
366: PetscErrorCode MatSetValuesRow_SeqAIJ(Mat A,PetscInt row,const PetscScalar v[])
367: {
368: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
369: PetscInt *ai = a->i;
373: PetscMemcpy(a->a+ai[row],v,(ai[row+1]-ai[row])*sizeof(PetscScalar));
374: return(0);
375: }
377: /*
378: MatSeqAIJSetValuesLocalFast - An optimized version of MatSetValuesLocal() for SeqAIJ matrices with several assumptions
380: - a single row of values is set with each call
381: - no row or column indices are negative or (in error) larger than the number of rows or columns
382: - the values are always added to the matrix, not set
383: - no new locations are introduced in the nonzero structure of the matrix
385: This does NOT assume the global column indices are sorted
387: */
389: #include <petsc-private/isimpl.h>
392: PetscErrorCode MatSeqAIJSetValuesLocalFast(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode is)
393: {
394: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
395: PetscInt low,high,t,row,nrow,i,col,l;
396: const PetscInt *rp,*ai = a->i,*ailen = a->ilen,*aj = a->j;
397: PetscInt lastcol = -1;
398: MatScalar *ap,value,*aa = a->a;
399: const PetscInt *ridx = A->rmap->mapping->indices,*cidx = A->cmap->mapping->indices;
401: row = ridx[im[0]];
402: rp = aj + ai[row];
403: ap = aa + ai[row];
404: nrow = ailen[row];
405: low = 0;
406: high = nrow;
407: for (l=0; l<n; l++) { /* loop over added columns */
408: col = cidx[in[l]];
409: value = v[l];
411: if (col <= lastcol) low = 0;
412: else high = nrow;
413: lastcol = col;
414: while (high-low > 5) {
415: t = (low+high)/2;
416: if (rp[t] > col) high = t;
417: else low = t;
418: }
419: for (i=low; i<high; i++) {
420: if (rp[i] == col) {
421: ap[i] += value;
422: low = i + 1;
423: break;
424: }
425: }
426: }
427: return 0;
428: }
432: PetscErrorCode MatSetValues_SeqAIJ(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode is)
433: {
434: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
435: PetscInt *rp,k,low,high,t,ii,row,nrow,i,col,l,rmax,N;
436: PetscInt *imax = a->imax,*ai = a->i,*ailen = a->ilen;
438: PetscInt *aj = a->j,nonew = a->nonew,lastcol = -1;
439: MatScalar *ap,value,*aa = a->a;
440: PetscBool ignorezeroentries = a->ignorezeroentries;
441: PetscBool roworiented = a->roworiented;
444: for (k=0; k<m; k++) { /* loop over added rows */
445: row = im[k];
446: if (row < 0) continue;
447: #if defined(PETSC_USE_DEBUG)
448: 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);
449: #endif
450: rp = aj + ai[row]; ap = aa + ai[row];
451: rmax = imax[row]; nrow = ailen[row];
452: low = 0;
453: high = nrow;
454: for (l=0; l<n; l++) { /* loop over added columns */
455: if (in[l] < 0) continue;
456: #if defined(PETSC_USE_DEBUG)
457: 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);
458: #endif
459: col = in[l];
460: if (roworiented) {
461: value = v[l + k*n];
462: } else {
463: value = v[k + l*m];
464: }
465: if ((value == 0.0 && ignorezeroentries) && (is == ADD_VALUES)) 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 (is == ADD_VALUES) ap[i] += value;
479: else ap[i] = value;
480: low = i + 1;
481: goto noinsert;
482: }
483: }
484: if (value == 0.0 && ignorezeroentries) goto noinsert;
485: if (nonew == 1) goto noinsert;
486: if (nonew == -1) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero at (%D,%D) in the matrix",row,col);
487: MatSeqXAIJReallocateAIJ(A,A->rmap->n,1,nrow,row,col,rmax,aa,ai,aj,rp,ap,imax,nonew,MatScalar);
488: N = nrow++ - 1; a->nz++; high++;
489: /* shift up all the later entries in this row */
490: for (ii=N; ii>=i; ii--) {
491: rp[ii+1] = rp[ii];
492: ap[ii+1] = ap[ii];
493: }
494: rp[i] = col;
495: ap[i] = value;
496: low = i + 1;
497: A->nonzerostate++;
498: noinsert:;
499: }
500: ailen[row] = nrow;
501: }
502: return(0);
503: }
508: PetscErrorCode MatGetValues_SeqAIJ(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],PetscScalar v[])
509: {
510: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
511: PetscInt *rp,k,low,high,t,row,nrow,i,col,l,*aj = a->j;
512: PetscInt *ai = a->i,*ailen = a->ilen;
513: MatScalar *ap,*aa = a->a;
516: for (k=0; k<m; k++) { /* loop over rows */
517: row = im[k];
518: if (row < 0) {v += n; continue;} /* SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative row: %D",row); */
519: 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);
520: rp = aj + ai[row]; ap = aa + ai[row];
521: nrow = ailen[row];
522: for (l=0; l<n; l++) { /* loop over columns */
523: if (in[l] < 0) {v++; continue;} /* SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative column: %D",in[l]); */
524: 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);
525: col = in[l];
526: high = nrow; low = 0; /* assume unsorted */
527: while (high-low > 5) {
528: t = (low+high)/2;
529: if (rp[t] > col) high = t;
530: else low = t;
531: }
532: for (i=low; i<high; i++) {
533: if (rp[i] > col) break;
534: if (rp[i] == col) {
535: *v++ = ap[i];
536: goto finished;
537: }
538: }
539: *v++ = 0.0;
540: finished:;
541: }
542: }
543: return(0);
544: }
549: PetscErrorCode MatView_SeqAIJ_Binary(Mat A,PetscViewer viewer)
550: {
551: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
553: PetscInt i,*col_lens;
554: int fd;
555: FILE *file;
558: PetscViewerBinaryGetDescriptor(viewer,&fd);
559: PetscMalloc1((4+A->rmap->n),&col_lens);
561: col_lens[0] = MAT_FILE_CLASSID;
562: col_lens[1] = A->rmap->n;
563: col_lens[2] = A->cmap->n;
564: col_lens[3] = a->nz;
566: /* store lengths of each row and write (including header) to file */
567: for (i=0; i<A->rmap->n; i++) {
568: col_lens[4+i] = a->i[i+1] - a->i[i];
569: }
570: PetscBinaryWrite(fd,col_lens,4+A->rmap->n,PETSC_INT,PETSC_TRUE);
571: PetscFree(col_lens);
573: /* store column indices (zero start index) */
574: PetscBinaryWrite(fd,a->j,a->nz,PETSC_INT,PETSC_FALSE);
576: /* store nonzero values */
577: PetscBinaryWrite(fd,a->a,a->nz,PETSC_SCALAR,PETSC_FALSE);
579: PetscViewerBinaryGetInfoPointer(viewer,&file);
580: if (file) {
581: fprintf(file,"-matload_block_size %d\n",(int)PetscAbs(A->rmap->bs));
582: }
583: return(0);
584: }
586: extern PetscErrorCode MatSeqAIJFactorInfo_Matlab(Mat,PetscViewer);
590: PetscErrorCode MatView_SeqAIJ_ASCII(Mat A,PetscViewer viewer)
591: {
592: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
593: PetscErrorCode ierr;
594: PetscInt i,j,m = A->rmap->n;
595: const char *name;
596: PetscViewerFormat format;
599: PetscViewerGetFormat(viewer,&format);
600: if (format == PETSC_VIEWER_ASCII_MATLAB) {
601: PetscInt nofinalvalue = 0;
602: if (m && ((a->i[m] == a->i[m-1]) || (a->j[a->nz-1] != A->cmap->n-1))) {
603: /* Need a dummy value to ensure the dimension of the matrix. */
604: nofinalvalue = 1;
605: }
606: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
607: PetscViewerASCIIPrintf(viewer,"%% Size = %D %D \n",m,A->cmap->n);
608: PetscViewerASCIIPrintf(viewer,"%% Nonzeros = %D \n",a->nz);
609: #if defined(PETSC_USE_COMPLEX)
610: PetscViewerASCIIPrintf(viewer,"zzz = zeros(%D,4);\n",a->nz+nofinalvalue);
611: #else
612: PetscViewerASCIIPrintf(viewer,"zzz = zeros(%D,3);\n",a->nz+nofinalvalue);
613: #endif
614: PetscViewerASCIIPrintf(viewer,"zzz = [\n");
616: for (i=0; i<m; i++) {
617: for (j=a->i[i]; j<a->i[i+1]; j++) {
618: #if defined(PETSC_USE_COMPLEX)
619: 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]));
620: #else
621: PetscViewerASCIIPrintf(viewer,"%D %D %18.16e\n",i+1,a->j[j]+1,(double)a->a[j]);
622: #endif
623: }
624: }
625: if (nofinalvalue) {
626: #if defined(PETSC_USE_COMPLEX)
627: PetscViewerASCIIPrintf(viewer,"%D %D %18.16e %18.16e\n",m,A->cmap->n,0.,0.);
628: #else
629: PetscViewerASCIIPrintf(viewer,"%D %D %18.16e\n",m,A->cmap->n,0.0);
630: #endif
631: }
632: PetscObjectGetName((PetscObject)A,&name);
633: PetscViewerASCIIPrintf(viewer,"];\n %s = spconvert(zzz);\n",name);
634: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
635: } else if (format == PETSC_VIEWER_ASCII_FACTOR_INFO || format == PETSC_VIEWER_ASCII_INFO) {
636: return(0);
637: } else if (format == PETSC_VIEWER_ASCII_COMMON) {
638: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
639: for (i=0; i<m; i++) {
640: PetscViewerASCIIPrintf(viewer,"row %D:",i);
641: for (j=a->i[i]; j<a->i[i+1]; j++) {
642: #if defined(PETSC_USE_COMPLEX)
643: if (PetscImaginaryPart(a->a[j]) > 0.0 && PetscRealPart(a->a[j]) != 0.0) {
644: PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));
645: } else if (PetscImaginaryPart(a->a[j]) < 0.0 && PetscRealPart(a->a[j]) != 0.0) {
646: PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)-PetscImaginaryPart(a->a[j]));
647: } else if (PetscRealPart(a->a[j]) != 0.0) {
648: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)PetscRealPart(a->a[j]));
649: }
650: #else
651: if (a->a[j] != 0.0) {PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)a->a[j]);}
652: #endif
653: }
654: PetscViewerASCIIPrintf(viewer,"\n");
655: }
656: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
657: } else if (format == PETSC_VIEWER_ASCII_SYMMODU) {
658: PetscInt nzd=0,fshift=1,*sptr;
659: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
660: PetscMalloc1((m+1),&sptr);
661: for (i=0; i<m; i++) {
662: sptr[i] = nzd+1;
663: for (j=a->i[i]; j<a->i[i+1]; j++) {
664: if (a->j[j] >= i) {
665: #if defined(PETSC_USE_COMPLEX)
666: if (PetscImaginaryPart(a->a[j]) != 0.0 || PetscRealPart(a->a[j]) != 0.0) nzd++;
667: #else
668: if (a->a[j] != 0.0) nzd++;
669: #endif
670: }
671: }
672: }
673: sptr[m] = nzd+1;
674: PetscViewerASCIIPrintf(viewer," %D %D\n\n",m,nzd);
675: for (i=0; i<m+1; i+=6) {
676: if (i+4<m) {
677: 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]);
678: } else if (i+3<m) {
679: PetscViewerASCIIPrintf(viewer," %D %D %D %D %D\n",sptr[i],sptr[i+1],sptr[i+2],sptr[i+3],sptr[i+4]);
680: } else if (i+2<m) {
681: PetscViewerASCIIPrintf(viewer," %D %D %D %D\n",sptr[i],sptr[i+1],sptr[i+2],sptr[i+3]);
682: } else if (i+1<m) {
683: PetscViewerASCIIPrintf(viewer," %D %D %D\n",sptr[i],sptr[i+1],sptr[i+2]);
684: } else if (i<m) {
685: PetscViewerASCIIPrintf(viewer," %D %D\n",sptr[i],sptr[i+1]);
686: } else {
687: PetscViewerASCIIPrintf(viewer," %D\n",sptr[i]);
688: }
689: }
690: PetscViewerASCIIPrintf(viewer,"\n");
691: PetscFree(sptr);
692: for (i=0; i<m; i++) {
693: for (j=a->i[i]; j<a->i[i+1]; j++) {
694: if (a->j[j] >= i) {PetscViewerASCIIPrintf(viewer," %D ",a->j[j]+fshift);}
695: }
696: PetscViewerASCIIPrintf(viewer,"\n");
697: }
698: PetscViewerASCIIPrintf(viewer,"\n");
699: for (i=0; i<m; i++) {
700: for (j=a->i[i]; j<a->i[i+1]; j++) {
701: if (a->j[j] >= i) {
702: #if defined(PETSC_USE_COMPLEX)
703: if (PetscImaginaryPart(a->a[j]) != 0.0 || PetscRealPart(a->a[j]) != 0.0) {
704: PetscViewerASCIIPrintf(viewer," %18.16e %18.16e ",(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));
705: }
706: #else
707: if (a->a[j] != 0.0) {PetscViewerASCIIPrintf(viewer," %18.16e ",(double)a->a[j]);}
708: #endif
709: }
710: }
711: PetscViewerASCIIPrintf(viewer,"\n");
712: }
713: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
714: } else if (format == PETSC_VIEWER_ASCII_DENSE) {
715: PetscInt cnt = 0,jcnt;
716: PetscScalar value;
717: #if defined(PETSC_USE_COMPLEX)
718: PetscBool realonly = PETSC_TRUE;
720: for (i=0; i<a->i[m]; i++) {
721: if (PetscImaginaryPart(a->a[i]) != 0.0) {
722: realonly = PETSC_FALSE;
723: break;
724: }
725: }
726: #endif
728: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
729: for (i=0; i<m; i++) {
730: jcnt = 0;
731: for (j=0; j<A->cmap->n; j++) {
732: if (jcnt < a->i[i+1]-a->i[i] && j == a->j[cnt]) {
733: value = a->a[cnt++];
734: jcnt++;
735: } else {
736: value = 0.0;
737: }
738: #if defined(PETSC_USE_COMPLEX)
739: if (realonly) {
740: PetscViewerASCIIPrintf(viewer," %7.5e ",(double)PetscRealPart(value));
741: } else {
742: PetscViewerASCIIPrintf(viewer," %7.5e+%7.5e i ",(double)PetscRealPart(value),(double)PetscImaginaryPart(value));
743: }
744: #else
745: PetscViewerASCIIPrintf(viewer," %7.5e ",(double)value);
746: #endif
747: }
748: PetscViewerASCIIPrintf(viewer,"\n");
749: }
750: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
751: } else if (format == PETSC_VIEWER_ASCII_MATRIXMARKET) {
752: PetscInt fshift=1;
753: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
754: #if defined(PETSC_USE_COMPLEX)
755: PetscViewerASCIIPrintf(viewer,"%%matrix complex general\n");
756: #else
757: PetscViewerASCIIPrintf(viewer,"%%matrix real general\n");
758: #endif
759: PetscViewerASCIIPrintf(viewer,"%D %D %D\n", m, A->cmap->n, a->nz);
760: for (i=0; i<m; i++) {
761: for (j=a->i[i]; j<a->i[i+1]; j++) {
762: #if defined(PETSC_USE_COMPLEX)
763: if (PetscImaginaryPart(a->a[j]) > 0.0) {
764: PetscViewerASCIIPrintf(viewer,"%D %D, %g %g\n", i+fshift,a->j[j]+fshift,(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));
765: } else if (PetscImaginaryPart(a->a[j]) < 0.0) {
766: PetscViewerASCIIPrintf(viewer,"%D %D, %g -%g\n", i+fshift,a->j[j]+fshift,(double)PetscRealPart(a->a[j]),(double)-PetscImaginaryPart(a->a[j]));
767: } else {
768: PetscViewerASCIIPrintf(viewer,"%D %D, %g\n", i+fshift,a->j[j]+fshift,(double)PetscRealPart(a->a[j]));
769: }
770: #else
771: PetscViewerASCIIPrintf(viewer,"%D %D %g\n", i+fshift, a->j[j]+fshift, (double)a->a[j]);
772: #endif
773: }
774: }
775: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
776: } else {
777: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
778: if (A->factortype) {
779: for (i=0; i<m; i++) {
780: PetscViewerASCIIPrintf(viewer,"row %D:",i);
781: /* L part */
782: for (j=a->i[i]; j<a->i[i+1]; j++) {
783: #if defined(PETSC_USE_COMPLEX)
784: if (PetscImaginaryPart(a->a[j]) > 0.0) {
785: PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));
786: } else if (PetscImaginaryPart(a->a[j]) < 0.0) {
787: PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)(-PetscImaginaryPart(a->a[j])));
788: } else {
789: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)PetscRealPart(a->a[j]));
790: }
791: #else
792: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)a->a[j]);
793: #endif
794: }
795: /* diagonal */
796: j = a->diag[i];
797: #if defined(PETSC_USE_COMPLEX)
798: if (PetscImaginaryPart(a->a[j]) > 0.0) {
799: PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j],(double)PetscRealPart(1.0/a->a[j]),(double)PetscImaginaryPart(1.0/a->a[j]));
800: } else if (PetscImaginaryPart(a->a[j]) < 0.0) {
801: PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j],(double)PetscRealPart(1.0/a->a[j]),(double)(-PetscImaginaryPart(1.0/a->a[j])));
802: } else {
803: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)PetscRealPart(1.0/a->a[j]));
804: }
805: #else
806: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)(1.0/a->a[j]));
807: #endif
809: /* U part */
810: for (j=a->diag[i+1]+1; j<a->diag[i]; j++) {
811: #if defined(PETSC_USE_COMPLEX)
812: if (PetscImaginaryPart(a->a[j]) > 0.0) {
813: PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));
814: } else if (PetscImaginaryPart(a->a[j]) < 0.0) {
815: PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)(-PetscImaginaryPart(a->a[j])));
816: } else {
817: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)PetscRealPart(a->a[j]));
818: }
819: #else
820: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)a->a[j]);
821: #endif
822: }
823: PetscViewerASCIIPrintf(viewer,"\n");
824: }
825: } else {
826: for (i=0; i<m; i++) {
827: PetscViewerASCIIPrintf(viewer,"row %D:",i);
828: for (j=a->i[i]; j<a->i[i+1]; j++) {
829: #if defined(PETSC_USE_COMPLEX)
830: if (PetscImaginaryPart(a->a[j]) > 0.0) {
831: PetscViewerASCIIPrintf(viewer," (%D, %g + %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)PetscImaginaryPart(a->a[j]));
832: } else if (PetscImaginaryPart(a->a[j]) < 0.0) {
833: PetscViewerASCIIPrintf(viewer," (%D, %g - %g i)",a->j[j],(double)PetscRealPart(a->a[j]),(double)-PetscImaginaryPart(a->a[j]));
834: } else {
835: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)PetscRealPart(a->a[j]));
836: }
837: #else
838: PetscViewerASCIIPrintf(viewer," (%D, %g) ",a->j[j],(double)a->a[j]);
839: #endif
840: }
841: PetscViewerASCIIPrintf(viewer,"\n");
842: }
843: }
844: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
845: }
846: PetscViewerFlush(viewer);
847: return(0);
848: }
850: #include <petscdraw.h>
853: PetscErrorCode MatView_SeqAIJ_Draw_Zoom(PetscDraw draw,void *Aa)
854: {
855: Mat A = (Mat) Aa;
856: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
857: PetscErrorCode ierr;
858: PetscInt i,j,m = A->rmap->n,color;
859: PetscReal xl,yl,xr,yr,x_l,x_r,y_l,y_r,maxv = 0.0;
860: PetscViewer viewer;
861: PetscViewerFormat format;
864: PetscObjectQuery((PetscObject)A,"Zoomviewer",(PetscObject*)&viewer);
865: PetscViewerGetFormat(viewer,&format);
867: PetscDrawGetCoordinates(draw,&xl,&yl,&xr,&yr);
868: /* loop over matrix elements drawing boxes */
870: if (format != PETSC_VIEWER_DRAW_CONTOUR) {
871: /* Blue for negative, Cyan for zero and Red for positive */
872: color = PETSC_DRAW_BLUE;
873: for (i=0; i<m; i++) {
874: y_l = m - i - 1.0; y_r = y_l + 1.0;
875: for (j=a->i[i]; j<a->i[i+1]; j++) {
876: x_l = a->j[j]; x_r = x_l + 1.0;
877: if (PetscRealPart(a->a[j]) >= 0.) continue;
878: PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);
879: }
880: }
881: color = PETSC_DRAW_CYAN;
882: for (i=0; i<m; i++) {
883: y_l = m - i - 1.0; y_r = y_l + 1.0;
884: for (j=a->i[i]; j<a->i[i+1]; j++) {
885: x_l = a->j[j]; x_r = x_l + 1.0;
886: if (a->a[j] != 0.) continue;
887: PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);
888: }
889: }
890: color = PETSC_DRAW_RED;
891: for (i=0; i<m; i++) {
892: y_l = m - i - 1.0; y_r = y_l + 1.0;
893: for (j=a->i[i]; j<a->i[i+1]; j++) {
894: x_l = a->j[j]; x_r = x_l + 1.0;
895: if (PetscRealPart(a->a[j]) <= 0.) continue;
896: PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);
897: }
898: }
899: } else {
900: /* use contour shading to indicate magnitude of values */
901: /* first determine max of all nonzero values */
902: PetscInt nz = a->nz,count;
903: PetscDraw popup;
904: PetscReal scale;
906: for (i=0; i<nz; i++) {
907: if (PetscAbsScalar(a->a[i]) > maxv) maxv = PetscAbsScalar(a->a[i]);
908: }
909: scale = (245.0 - PETSC_DRAW_BASIC_COLORS)/maxv;
910: PetscDrawGetPopup(draw,&popup);
911: if (popup) {
912: PetscDrawScalePopup(popup,0.0,maxv);
913: }
914: count = 0;
915: for (i=0; i<m; i++) {
916: y_l = m - i - 1.0; y_r = y_l + 1.0;
917: for (j=a->i[i]; j<a->i[i+1]; j++) {
918: x_l = a->j[j]; x_r = x_l + 1.0;
919: color = PETSC_DRAW_BASIC_COLORS + (PetscInt)(scale*PetscAbsScalar(a->a[count]));
920: PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);
921: count++;
922: }
923: }
924: }
925: return(0);
926: }
928: #include <petscdraw.h>
931: PetscErrorCode MatView_SeqAIJ_Draw(Mat A,PetscViewer viewer)
932: {
934: PetscDraw draw;
935: PetscReal xr,yr,xl,yl,h,w;
936: PetscBool isnull;
939: PetscViewerDrawGetDraw(viewer,0,&draw);
940: PetscDrawIsNull(draw,&isnull);
941: if (isnull) return(0);
943: PetscObjectCompose((PetscObject)A,"Zoomviewer",(PetscObject)viewer);
944: xr = A->cmap->n; yr = A->rmap->n; h = yr/10.0; w = xr/10.0;
945: xr += w; yr += h; xl = -w; yl = -h;
946: PetscDrawSetCoordinates(draw,xl,yl,xr,yr);
947: PetscDrawZoom(draw,MatView_SeqAIJ_Draw_Zoom,A);
948: PetscObjectCompose((PetscObject)A,"Zoomviewer",NULL);
949: return(0);
950: }
954: PetscErrorCode MatView_SeqAIJ(Mat A,PetscViewer viewer)
955: {
957: PetscBool iascii,isbinary,isdraw;
960: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);
961: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary);
962: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERDRAW,&isdraw);
963: if (iascii) {
964: MatView_SeqAIJ_ASCII(A,viewer);
965: } else if (isbinary) {
966: MatView_SeqAIJ_Binary(A,viewer);
967: } else if (isdraw) {
968: MatView_SeqAIJ_Draw(A,viewer);
969: }
970: MatView_SeqAIJ_Inode(A,viewer);
971: return(0);
972: }
976: PetscErrorCode MatAssemblyEnd_SeqAIJ(Mat A,MatAssemblyType mode)
977: {
978: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
980: PetscInt fshift = 0,i,j,*ai = a->i,*aj = a->j,*imax = a->imax;
981: PetscInt m = A->rmap->n,*ip,N,*ailen = a->ilen,rmax = 0;
982: MatScalar *aa = a->a,*ap;
983: PetscReal ratio = 0.6;
986: if (mode == MAT_FLUSH_ASSEMBLY) return(0);
988: if (m) rmax = ailen[0]; /* determine row with most nonzeros */
989: for (i=1; i<m; i++) {
990: /* move each row back by the amount of empty slots (fshift) before it*/
991: fshift += imax[i-1] - ailen[i-1];
992: rmax = PetscMax(rmax,ailen[i]);
993: if (fshift) {
994: ip = aj + ai[i];
995: ap = aa + ai[i];
996: N = ailen[i];
997: for (j=0; j<N; j++) {
998: ip[j-fshift] = ip[j];
999: ap[j-fshift] = ap[j];
1000: }
1001: }
1002: ai[i] = ai[i-1] + ailen[i-1];
1003: }
1004: if (m) {
1005: fshift += imax[m-1] - ailen[m-1];
1006: ai[m] = ai[m-1] + ailen[m-1];
1007: }
1009: /* reset ilen and imax for each row */
1010: a->nonzerorowcnt = 0;
1011: for (i=0; i<m; i++) {
1012: ailen[i] = imax[i] = ai[i+1] - ai[i];
1013: a->nonzerorowcnt += ((ai[i+1] - ai[i]) > 0);
1014: }
1015: a->nz = ai[m];
1016: 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);
1018: MatMarkDiagonal_SeqAIJ(A);
1019: PetscInfo4(A,"Matrix size: %D X %D; storage space: %D unneeded,%D used\n",m,A->cmap->n,fshift,a->nz);
1020: PetscInfo1(A,"Number of mallocs during MatSetValues() is %D\n",a->reallocs);
1021: PetscInfo1(A,"Maximum nonzeros in any row is %D\n",rmax);
1023: A->info.mallocs += a->reallocs;
1024: a->reallocs = 0;
1025: A->info.nz_unneeded = (PetscReal)fshift;
1026: a->rmax = rmax;
1028: MatCheckCompressedRow(A,a->nonzerorowcnt,&a->compressedrow,a->i,m,ratio);
1029: MatAssemblyEnd_SeqAIJ_Inode(A,mode);
1030: MatSeqAIJInvalidateDiagonal(A);
1031: return(0);
1032: }
1036: PetscErrorCode MatRealPart_SeqAIJ(Mat A)
1037: {
1038: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1039: PetscInt i,nz = a->nz;
1040: MatScalar *aa = a->a;
1044: for (i=0; i<nz; i++) aa[i] = PetscRealPart(aa[i]);
1045: MatSeqAIJInvalidateDiagonal(A);
1046: return(0);
1047: }
1051: PetscErrorCode MatImaginaryPart_SeqAIJ(Mat A)
1052: {
1053: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1054: PetscInt i,nz = a->nz;
1055: MatScalar *aa = a->a;
1059: for (i=0; i<nz; i++) aa[i] = PetscImaginaryPart(aa[i]);
1060: MatSeqAIJInvalidateDiagonal(A);
1061: return(0);
1062: }
1064: #if defined(PETSC_THREADCOMM_ACTIVE)
1065: PetscErrorCode MatZeroEntries_SeqAIJ_Kernel(PetscInt thread_id,Mat A)
1066: {
1068: PetscInt *trstarts=A->rmap->trstarts;
1069: PetscInt n,start,end;
1070: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1072: start = trstarts[thread_id];
1073: end = trstarts[thread_id+1];
1074: n = a->i[end] - a->i[start];
1075: PetscMemzero(a->a+a->i[start],n*sizeof(PetscScalar));
1076: return 0;
1077: }
1081: PetscErrorCode MatZeroEntries_SeqAIJ(Mat A)
1082: {
1086: PetscThreadCommRunKernel(PetscObjectComm((PetscObject)A),(PetscThreadKernel)MatZeroEntries_SeqAIJ_Kernel,1,A);
1087: MatSeqAIJInvalidateDiagonal(A);
1088: return(0);
1089: }
1090: #else
1093: PetscErrorCode MatZeroEntries_SeqAIJ(Mat A)
1094: {
1095: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1099: PetscMemzero(a->a,(a->i[A->rmap->n])*sizeof(PetscScalar));
1100: MatSeqAIJInvalidateDiagonal(A);
1101: return(0);
1102: }
1103: #endif
1105: extern PetscErrorCode MatDestroy_Redundant(Mat_Redundant **);
1109: PetscErrorCode MatDestroy_SeqAIJ(Mat A)
1110: {
1111: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1115: #if defined(PETSC_USE_LOG)
1116: PetscLogObjectState((PetscObject)A,"Rows=%D, Cols=%D, NZ=%D",A->rmap->n,A->cmap->n,a->nz);
1117: #endif
1118: MatDestroy_Redundant(&a->redundant);
1119: MatSeqXAIJFreeAIJ(A,&a->a,&a->j,&a->i);
1120: ISDestroy(&a->row);
1121: ISDestroy(&a->col);
1122: PetscFree(a->diag);
1123: PetscFree(a->ibdiag);
1124: PetscFree2(a->imax,a->ilen);
1125: PetscFree3(a->idiag,a->mdiag,a->ssor_work);
1126: PetscFree(a->solve_work);
1127: ISDestroy(&a->icol);
1128: PetscFree(a->saved_values);
1129: ISColoringDestroy(&a->coloring);
1130: PetscFree(a->xtoy);
1131: MatDestroy(&a->XtoY);
1132: PetscFree2(a->compressedrow.i,a->compressedrow.rindex);
1133: PetscFree(a->matmult_abdense);
1135: MatDestroy_SeqAIJ_Inode(A);
1136: PetscFree(A->data);
1138: PetscObjectChangeTypeName((PetscObject)A,0);
1139: PetscObjectComposeFunction((PetscObject)A,"MatSeqAIJSetColumnIndices_C",NULL);
1140: PetscObjectComposeFunction((PetscObject)A,"MatStoreValues_C",NULL);
1141: PetscObjectComposeFunction((PetscObject)A,"MatRetrieveValues_C",NULL);
1142: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqsbaij_C",NULL);
1143: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqbaij_C",NULL);
1144: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqaij_seqaijperm_C",NULL);
1145: PetscObjectComposeFunction((PetscObject)A,"MatIsTranspose_C",NULL);
1146: PetscObjectComposeFunction((PetscObject)A,"MatSeqAIJSetPreallocation_C",NULL);
1147: PetscObjectComposeFunction((PetscObject)A,"MatSeqAIJSetPreallocationCSR_C",NULL);
1148: PetscObjectComposeFunction((PetscObject)A,"MatReorderForNonzeroDiagonal_C",NULL);
1149: return(0);
1150: }
1154: PetscErrorCode MatSetOption_SeqAIJ(Mat A,MatOption op,PetscBool flg)
1155: {
1156: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1160: switch (op) {
1161: case MAT_ROW_ORIENTED:
1162: a->roworiented = flg;
1163: break;
1164: case MAT_KEEP_NONZERO_PATTERN:
1165: a->keepnonzeropattern = flg;
1166: break;
1167: case MAT_NEW_NONZERO_LOCATIONS:
1168: a->nonew = (flg ? 0 : 1);
1169: break;
1170: case MAT_NEW_NONZERO_LOCATION_ERR:
1171: a->nonew = (flg ? -1 : 0);
1172: break;
1173: case MAT_NEW_NONZERO_ALLOCATION_ERR:
1174: a->nonew = (flg ? -2 : 0);
1175: break;
1176: case MAT_UNUSED_NONZERO_LOCATION_ERR:
1177: a->nounused = (flg ? -1 : 0);
1178: break;
1179: case MAT_IGNORE_ZERO_ENTRIES:
1180: a->ignorezeroentries = flg;
1181: break;
1182: case MAT_SPD:
1183: case MAT_SYMMETRIC:
1184: case MAT_STRUCTURALLY_SYMMETRIC:
1185: case MAT_HERMITIAN:
1186: case MAT_SYMMETRY_ETERNAL:
1187: /* These options are handled directly by MatSetOption() */
1188: break;
1189: case MAT_NEW_DIAGONALS:
1190: case MAT_IGNORE_OFF_PROC_ENTRIES:
1191: case MAT_USE_HASH_TABLE:
1192: PetscInfo1(A,"Option %s ignored\n",MatOptions[op]);
1193: break;
1194: case MAT_USE_INODES:
1195: /* Not an error because MatSetOption_SeqAIJ_Inode handles this one */
1196: break;
1197: default:
1198: SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_SUP,"unknown option %d",op);
1199: }
1200: MatSetOption_SeqAIJ_Inode(A,op,flg);
1201: return(0);
1202: }
1206: PetscErrorCode MatGetDiagonal_SeqAIJ(Mat A,Vec v)
1207: {
1208: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1210: PetscInt i,j,n,*ai=a->i,*aj=a->j,nz;
1211: PetscScalar *aa=a->a,*x,zero=0.0;
1214: VecGetLocalSize(v,&n);
1215: if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector");
1217: if (A->factortype == MAT_FACTOR_ILU || A->factortype == MAT_FACTOR_LU) {
1218: PetscInt *diag=a->diag;
1219: VecGetArray(v,&x);
1220: for (i=0; i<n; i++) x[i] = 1.0/aa[diag[i]];
1221: VecRestoreArray(v,&x);
1222: return(0);
1223: }
1225: VecSet(v,zero);
1226: VecGetArray(v,&x);
1227: for (i=0; i<n; i++) {
1228: nz = ai[i+1] - ai[i];
1229: if (!nz) x[i] = 0.0;
1230: for (j=ai[i]; j<ai[i+1]; j++) {
1231: if (aj[j] == i) {
1232: x[i] = aa[j];
1233: break;
1234: }
1235: }
1236: }
1237: VecRestoreArray(v,&x);
1238: return(0);
1239: }
1241: #include <../src/mat/impls/aij/seq/ftn-kernels/fmult.h>
1244: PetscErrorCode MatMultTransposeAdd_SeqAIJ(Mat A,Vec xx,Vec zz,Vec yy)
1245: {
1246: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1247: PetscScalar *x,*y;
1249: PetscInt m = A->rmap->n;
1250: #if !defined(PETSC_USE_FORTRAN_KERNEL_MULTTRANSPOSEAIJ)
1251: MatScalar *v;
1252: PetscScalar alpha;
1253: PetscInt n,i,j,*idx,*ii,*ridx=NULL;
1254: Mat_CompressedRow cprow = a->compressedrow;
1255: PetscBool usecprow = cprow.use;
1256: #endif
1259: if (zz != yy) {VecCopy(zz,yy);}
1260: VecGetArray(xx,&x);
1261: VecGetArray(yy,&y);
1263: #if defined(PETSC_USE_FORTRAN_KERNEL_MULTTRANSPOSEAIJ)
1264: fortranmulttransposeaddaij_(&m,x,a->i,a->j,a->a,y);
1265: #else
1266: if (usecprow) {
1267: m = cprow.nrows;
1268: ii = cprow.i;
1269: ridx = cprow.rindex;
1270: } else {
1271: ii = a->i;
1272: }
1273: for (i=0; i<m; i++) {
1274: idx = a->j + ii[i];
1275: v = a->a + ii[i];
1276: n = ii[i+1] - ii[i];
1277: if (usecprow) {
1278: alpha = x[ridx[i]];
1279: } else {
1280: alpha = x[i];
1281: }
1282: for (j=0; j<n; j++) y[idx[j]] += alpha*v[j];
1283: }
1284: #endif
1285: PetscLogFlops(2.0*a->nz);
1286: VecRestoreArray(xx,&x);
1287: VecRestoreArray(yy,&y);
1288: return(0);
1289: }
1293: PetscErrorCode MatMultTranspose_SeqAIJ(Mat A,Vec xx,Vec yy)
1294: {
1298: VecSet(yy,0.0);
1299: MatMultTransposeAdd_SeqAIJ(A,xx,yy,yy);
1300: return(0);
1301: }
1303: #include <../src/mat/impls/aij/seq/ftn-kernels/fmult.h>
1304: #if defined(PETSC_THREADCOMM_ACTIVE)
1305: PetscErrorCode MatMult_SeqAIJ_Kernel(PetscInt thread_id,Mat A,Vec xx,Vec yy)
1306: {
1307: PetscErrorCode ierr;
1308: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1309: PetscScalar *y;
1310: const PetscScalar *x;
1311: const MatScalar *aa;
1312: PetscInt *trstarts=A->rmap->trstarts;
1313: PetscInt n,start,end,i;
1314: const PetscInt *aj,*ai;
1315: PetscScalar sum;
1317: VecGetArrayRead(xx,&x);
1318: VecGetArray(yy,&y);
1319: start = trstarts[thread_id];
1320: end = trstarts[thread_id+1];
1321: aj = a->j;
1322: aa = a->a;
1323: ai = a->i;
1324: for (i=start; i<end; i++) {
1325: n = ai[i+1] - ai[i];
1326: aj = a->j + ai[i];
1327: aa = a->a + ai[i];
1328: sum = 0.0;
1329: PetscSparseDensePlusDot(sum,x,aa,aj,n);
1330: y[i] = sum;
1331: }
1332: VecRestoreArrayRead(xx,&x);
1333: VecRestoreArray(yy,&y);
1334: return 0;
1335: }
1339: PetscErrorCode MatMult_SeqAIJ(Mat A,Vec xx,Vec yy)
1340: {
1341: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1342: PetscScalar *y;
1343: const PetscScalar *x;
1344: const MatScalar *aa;
1345: PetscErrorCode ierr;
1346: PetscInt m=A->rmap->n;
1347: const PetscInt *aj,*ii,*ridx=NULL;
1348: PetscInt n,i;
1349: PetscScalar sum;
1350: PetscBool usecprow=a->compressedrow.use;
1352: #if defined(PETSC_HAVE_PRAGMA_DISJOINT)
1353: #pragma disjoint(*x,*y,*aa)
1354: #endif
1357: aj = a->j;
1358: aa = a->a;
1359: ii = a->i;
1360: if (usecprow) { /* use compressed row format */
1361: VecGetArrayRead(xx,&x);
1362: VecGetArray(yy,&y);
1363: PetscMemzero(y,m*sizeof(PetscScalar));
1364: m = a->compressedrow.nrows;
1365: ii = a->compressedrow.i;
1366: ridx = a->compressedrow.rindex;
1367: for (i=0; i<m; i++) {
1368: n = ii[i+1] - ii[i];
1369: aj = a->j + ii[i];
1370: aa = a->a + ii[i];
1371: sum = 0.0;
1372: PetscSparseDensePlusDot(sum,x,aa,aj,n);
1373: /* for (j=0; j<n; j++) sum += (*aa++)*x[*aj++]; */
1374: y[*ridx++] = sum;
1375: }
1376: VecRestoreArrayRead(xx,&x);
1377: VecRestoreArray(yy,&y);
1378: } else { /* do not use compressed row format */
1379: #if defined(PETSC_USE_FORTRAN_KERNEL_MULTAIJ)
1380: fortranmultaij_(&m,x,ii,aj,aa,y);
1381: #else
1382: PetscThreadCommRunKernel(PetscObjectComm((PetscObject)A),(PetscThreadKernel)MatMult_SeqAIJ_Kernel,3,A,xx,yy);
1383: #endif
1384: }
1385: PetscLogFlops(2.0*a->nz - a->nonzerorowcnt);
1386: return(0);
1387: }
1388: #else
1391: PetscErrorCode MatMult_SeqAIJ(Mat A,Vec xx,Vec yy)
1392: {
1393: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1394: PetscScalar *y;
1395: const PetscScalar *x;
1396: const MatScalar *aa;
1397: PetscErrorCode ierr;
1398: PetscInt m=A->rmap->n;
1399: const PetscInt *aj,*ii,*ridx=NULL;
1400: PetscInt n,i;
1401: PetscScalar sum;
1402: PetscBool usecprow=a->compressedrow.use;
1404: #if defined(PETSC_HAVE_PRAGMA_DISJOINT)
1405: #pragma disjoint(*x,*y,*aa)
1406: #endif
1409: VecGetArrayRead(xx,&x);
1410: VecGetArray(yy,&y);
1411: aj = a->j;
1412: aa = a->a;
1413: ii = a->i;
1414: if (usecprow) { /* use compressed row format */
1415: PetscMemzero(y,m*sizeof(PetscScalar));
1416: m = a->compressedrow.nrows;
1417: ii = a->compressedrow.i;
1418: ridx = a->compressedrow.rindex;
1419: for (i=0; i<m; i++) {
1420: n = ii[i+1] - ii[i];
1421: aj = a->j + ii[i];
1422: aa = a->a + ii[i];
1423: sum = 0.0;
1424: PetscSparseDensePlusDot(sum,x,aa,aj,n);
1425: /* for (j=0; j<n; j++) sum += (*aa++)*x[*aj++]; */
1426: y[*ridx++] = sum;
1427: }
1428: } else { /* do not use compressed row format */
1429: #if defined(PETSC_USE_FORTRAN_KERNEL_MULTAIJ)
1430: fortranmultaij_(&m,x,ii,aj,aa,y);
1431: #else
1432: #if defined(PETSC_THREADCOMM_ACTIVE)
1433: PetscThreadCommRunKernel(PetscObjectComm((PetscObject)A),(PetscThreadKernel)MatMult_SeqAIJ_Kernel,3,A,xx,yy);
1434: #else
1435: for (i=0; i<m; i++) {
1436: n = ii[i+1] - ii[i];
1437: aj = a->j + ii[i];
1438: aa = a->a + ii[i];
1439: sum = 0.0;
1440: PetscSparseDensePlusDot(sum,x,aa,aj,n);
1441: y[i] = sum;
1442: }
1443: #endif
1444: #endif
1445: }
1446: PetscLogFlops(2.0*a->nz - a->nonzerorowcnt);
1447: VecRestoreArrayRead(xx,&x);
1448: VecRestoreArray(yy,&y);
1449: return(0);
1450: }
1451: #endif
1455: PetscErrorCode MatMultMax_SeqAIJ(Mat A,Vec xx,Vec yy)
1456: {
1457: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1458: PetscScalar *y;
1459: const PetscScalar *x;
1460: const MatScalar *aa;
1461: PetscErrorCode ierr;
1462: PetscInt m=A->rmap->n;
1463: const PetscInt *aj,*ii,*ridx=NULL;
1464: PetscInt n,i,nonzerorow=0;
1465: PetscScalar sum;
1466: PetscBool usecprow=a->compressedrow.use;
1468: #if defined(PETSC_HAVE_PRAGMA_DISJOINT)
1469: #pragma disjoint(*x,*y,*aa)
1470: #endif
1473: VecGetArrayRead(xx,&x);
1474: VecGetArray(yy,&y);
1475: aj = a->j;
1476: aa = a->a;
1477: ii = a->i;
1478: if (usecprow) { /* use compressed row format */
1479: m = a->compressedrow.nrows;
1480: ii = a->compressedrow.i;
1481: ridx = a->compressedrow.rindex;
1482: for (i=0; i<m; i++) {
1483: n = ii[i+1] - ii[i];
1484: aj = a->j + ii[i];
1485: aa = a->a + ii[i];
1486: sum = 0.0;
1487: nonzerorow += (n>0);
1488: PetscSparseDenseMaxDot(sum,x,aa,aj,n);
1489: /* for (j=0; j<n; j++) sum += (*aa++)*x[*aj++]; */
1490: y[*ridx++] = sum;
1491: }
1492: } else { /* do not use compressed row format */
1493: for (i=0; i<m; i++) {
1494: n = ii[i+1] - ii[i];
1495: aj = a->j + ii[i];
1496: aa = a->a + ii[i];
1497: sum = 0.0;
1498: nonzerorow += (n>0);
1499: PetscSparseDenseMaxDot(sum,x,aa,aj,n);
1500: y[i] = sum;
1501: }
1502: }
1503: PetscLogFlops(2.0*a->nz - nonzerorow);
1504: VecRestoreArrayRead(xx,&x);
1505: VecRestoreArray(yy,&y);
1506: return(0);
1507: }
1511: PetscErrorCode MatMultAddMax_SeqAIJ(Mat A,Vec xx,Vec yy,Vec zz)
1512: {
1513: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1514: PetscScalar *y,*z;
1515: const PetscScalar *x;
1516: const MatScalar *aa;
1517: PetscErrorCode ierr;
1518: PetscInt m = A->rmap->n,*aj,*ii;
1519: PetscInt n,i,*ridx=NULL;
1520: PetscScalar sum;
1521: PetscBool usecprow=a->compressedrow.use;
1524: VecGetArrayRead(xx,&x);
1525: VecGetArray(yy,&y);
1526: if (zz != yy) {
1527: VecGetArray(zz,&z);
1528: } else {
1529: z = y;
1530: }
1532: aj = a->j;
1533: aa = a->a;
1534: ii = a->i;
1535: if (usecprow) { /* use compressed row format */
1536: if (zz != yy) {
1537: PetscMemcpy(z,y,m*sizeof(PetscScalar));
1538: }
1539: m = a->compressedrow.nrows;
1540: ii = a->compressedrow.i;
1541: ridx = a->compressedrow.rindex;
1542: for (i=0; i<m; i++) {
1543: n = ii[i+1] - ii[i];
1544: aj = a->j + ii[i];
1545: aa = a->a + ii[i];
1546: sum = y[*ridx];
1547: PetscSparseDenseMaxDot(sum,x,aa,aj,n);
1548: z[*ridx++] = sum;
1549: }
1550: } else { /* do not use compressed row format */
1551: for (i=0; i<m; i++) {
1552: n = ii[i+1] - ii[i];
1553: aj = a->j + ii[i];
1554: aa = a->a + ii[i];
1555: sum = y[i];
1556: PetscSparseDenseMaxDot(sum,x,aa,aj,n);
1557: z[i] = sum;
1558: }
1559: }
1560: PetscLogFlops(2.0*a->nz);
1561: VecRestoreArrayRead(xx,&x);
1562: VecRestoreArray(yy,&y);
1563: if (zz != yy) {
1564: VecRestoreArray(zz,&z);
1565: }
1566: return(0);
1567: }
1569: #include <../src/mat/impls/aij/seq/ftn-kernels/fmultadd.h>
1572: PetscErrorCode MatMultAdd_SeqAIJ(Mat A,Vec xx,Vec yy,Vec zz)
1573: {
1574: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1575: PetscScalar *y,*z;
1576: const PetscScalar *x;
1577: const MatScalar *aa;
1578: PetscErrorCode ierr;
1579: PetscInt m = A->rmap->n,*aj,*ii;
1580: PetscInt n,i,*ridx=NULL;
1581: PetscScalar sum;
1582: PetscBool usecprow=a->compressedrow.use;
1585: VecGetArrayRead(xx,&x);
1586: VecGetArray(yy,&y);
1587: if (zz != yy) {
1588: VecGetArray(zz,&z);
1589: } else {
1590: z = y;
1591: }
1593: aj = a->j;
1594: aa = a->a;
1595: ii = a->i;
1596: if (usecprow) { /* use compressed row format */
1597: if (zz != yy) {
1598: PetscMemcpy(z,y,m*sizeof(PetscScalar));
1599: }
1600: m = a->compressedrow.nrows;
1601: ii = a->compressedrow.i;
1602: ridx = a->compressedrow.rindex;
1603: for (i=0; i<m; i++) {
1604: n = ii[i+1] - ii[i];
1605: aj = a->j + ii[i];
1606: aa = a->a + ii[i];
1607: sum = y[*ridx];
1608: PetscSparseDensePlusDot(sum,x,aa,aj,n);
1609: z[*ridx++] = sum;
1610: }
1611: } else { /* do not use compressed row format */
1612: #if defined(PETSC_USE_FORTRAN_KERNEL_MULTADDAIJ)
1613: fortranmultaddaij_(&m,x,ii,aj,aa,y,z);
1614: #else
1615: for (i=0; i<m; i++) {
1616: n = ii[i+1] - ii[i];
1617: aj = a->j + ii[i];
1618: aa = a->a + ii[i];
1619: sum = y[i];
1620: PetscSparseDensePlusDot(sum,x,aa,aj,n);
1621: z[i] = sum;
1622: }
1623: #endif
1624: }
1625: PetscLogFlops(2.0*a->nz);
1626: VecRestoreArrayRead(xx,&x);
1627: VecRestoreArray(yy,&y);
1628: if (zz != yy) {
1629: VecRestoreArray(zz,&z);
1630: }
1631: #if defined(PETSC_HAVE_CUSP)
1632: /*
1633: VecView(xx,0);
1634: VecView(zz,0);
1635: MatView(A,0);
1636: */
1637: #endif
1638: return(0);
1639: }
1641: /*
1642: Adds diagonal pointers to sparse matrix structure.
1643: */
1646: PetscErrorCode MatMarkDiagonal_SeqAIJ(Mat A)
1647: {
1648: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1650: PetscInt i,j,m = A->rmap->n;
1653: if (!a->diag) {
1654: PetscMalloc1(m,&a->diag);
1655: PetscLogObjectMemory((PetscObject)A, m*sizeof(PetscInt));
1656: }
1657: for (i=0; i<A->rmap->n; i++) {
1658: a->diag[i] = a->i[i+1];
1659: for (j=a->i[i]; j<a->i[i+1]; j++) {
1660: if (a->j[j] == i) {
1661: a->diag[i] = j;
1662: break;
1663: }
1664: }
1665: }
1666: return(0);
1667: }
1669: /*
1670: Checks for missing diagonals
1671: */
1674: PetscErrorCode MatMissingDiagonal_SeqAIJ(Mat A,PetscBool *missing,PetscInt *d)
1675: {
1676: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1677: PetscInt *diag,*ii = a->i,i;
1680: *missing = PETSC_FALSE;
1681: if (A->rmap->n > 0 && !ii) {
1682: *missing = PETSC_TRUE;
1683: if (d) *d = 0;
1684: PetscInfo(A,"Matrix has no entries therefore is missing diagonal");
1685: } else {
1686: diag = a->diag;
1687: for (i=0; i<A->rmap->n; i++) {
1688: if (diag[i] >= ii[i+1]) {
1689: *missing = PETSC_TRUE;
1690: if (d) *d = i;
1691: PetscInfo1(A,"Matrix is missing diagonal number %D",i);
1692: break;
1693: }
1694: }
1695: }
1696: return(0);
1697: }
1701: PetscErrorCode MatInvertDiagonal_SeqAIJ(Mat A,PetscScalar omega,PetscScalar fshift)
1702: {
1703: Mat_SeqAIJ *a = (Mat_SeqAIJ*) A->data;
1705: PetscInt i,*diag,m = A->rmap->n;
1706: MatScalar *v = a->a;
1707: PetscScalar *idiag,*mdiag;
1710: if (a->idiagvalid) return(0);
1711: MatMarkDiagonal_SeqAIJ(A);
1712: diag = a->diag;
1713: if (!a->idiag) {
1714: PetscMalloc3(m,&a->idiag,m,&a->mdiag,m,&a->ssor_work);
1715: PetscLogObjectMemory((PetscObject)A, 3*m*sizeof(PetscScalar));
1716: v = a->a;
1717: }
1718: mdiag = a->mdiag;
1719: idiag = a->idiag;
1721: if (omega == 1.0 && !PetscAbsScalar(fshift)) {
1722: for (i=0; i<m; i++) {
1723: mdiag[i] = v[diag[i]];
1724: if (!PetscAbsScalar(mdiag[i])) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"Zero diagonal on row %D",i);
1725: idiag[i] = 1.0/v[diag[i]];
1726: }
1727: PetscLogFlops(m);
1728: } else {
1729: for (i=0; i<m; i++) {
1730: mdiag[i] = v[diag[i]];
1731: idiag[i] = omega/(fshift + v[diag[i]]);
1732: }
1733: PetscLogFlops(2.0*m);
1734: }
1735: a->idiagvalid = PETSC_TRUE;
1736: return(0);
1737: }
1739: #include <../src/mat/impls/aij/seq/ftn-kernels/frelax.h>
1742: PetscErrorCode MatSOR_SeqAIJ(Mat A,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,PetscInt its,PetscInt lits,Vec xx)
1743: {
1744: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1745: PetscScalar *x,d,sum,*t,scale;
1746: const MatScalar *v = a->a,*idiag=0,*mdiag;
1747: const PetscScalar *b, *bs,*xb, *ts;
1748: PetscErrorCode ierr;
1749: PetscInt n = A->cmap->n,m = A->rmap->n,i;
1750: const PetscInt *idx,*diag;
1753: its = its*lits;
1755: if (fshift != a->fshift || omega != a->omega) a->idiagvalid = PETSC_FALSE; /* must recompute idiag[] */
1756: if (!a->idiagvalid) {MatInvertDiagonal_SeqAIJ(A,omega,fshift);}
1757: a->fshift = fshift;
1758: a->omega = omega;
1760: diag = a->diag;
1761: t = a->ssor_work;
1762: idiag = a->idiag;
1763: mdiag = a->mdiag;
1765: VecGetArray(xx,&x);
1766: VecGetArrayRead(bb,&b);
1767: /* We count flops by assuming the upper triangular and lower triangular parts have the same number of nonzeros */
1768: if (flag == SOR_APPLY_UPPER) {
1769: /* apply (U + D/omega) to the vector */
1770: bs = b;
1771: for (i=0; i<m; i++) {
1772: d = fshift + mdiag[i];
1773: n = a->i[i+1] - diag[i] - 1;
1774: idx = a->j + diag[i] + 1;
1775: v = a->a + diag[i] + 1;
1776: sum = b[i]*d/omega;
1777: PetscSparseDensePlusDot(sum,bs,v,idx,n);
1778: x[i] = sum;
1779: }
1780: VecRestoreArray(xx,&x);
1781: VecRestoreArrayRead(bb,&b);
1782: PetscLogFlops(a->nz);
1783: return(0);
1784: }
1786: if (flag == SOR_APPLY_LOWER) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"SOR_APPLY_LOWER is not implemented");
1787: else if (flag & SOR_EISENSTAT) {
1788: /* Let A = L + U + D; where L is lower trianglar,
1789: U is upper triangular, E = D/omega; This routine applies
1791: (L + E)^{-1} A (U + E)^{-1}
1793: to a vector efficiently using Eisenstat's trick.
1794: */
1795: scale = (2.0/omega) - 1.0;
1797: /* x = (E + U)^{-1} b */
1798: for (i=m-1; i>=0; i--) {
1799: n = a->i[i+1] - diag[i] - 1;
1800: idx = a->j + diag[i] + 1;
1801: v = a->a + diag[i] + 1;
1802: sum = b[i];
1803: PetscSparseDenseMinusDot(sum,x,v,idx,n);
1804: x[i] = sum*idiag[i];
1805: }
1807: /* t = b - (2*E - D)x */
1808: v = a->a;
1809: for (i=0; i<m; i++) t[i] = b[i] - scale*(v[*diag++])*x[i];
1811: /* t = (E + L)^{-1}t */
1812: ts = t;
1813: diag = a->diag;
1814: for (i=0; i<m; i++) {
1815: n = diag[i] - a->i[i];
1816: idx = a->j + a->i[i];
1817: v = a->a + a->i[i];
1818: sum = t[i];
1819: PetscSparseDenseMinusDot(sum,ts,v,idx,n);
1820: t[i] = sum*idiag[i];
1821: /* x = x + t */
1822: x[i] += t[i];
1823: }
1825: PetscLogFlops(6.0*m-1 + 2.0*a->nz);
1826: VecRestoreArray(xx,&x);
1827: VecRestoreArrayRead(bb,&b);
1828: return(0);
1829: }
1830: if (flag & SOR_ZERO_INITIAL_GUESS) {
1831: if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) {
1832: for (i=0; i<m; i++) {
1833: n = diag[i] - a->i[i];
1834: idx = a->j + a->i[i];
1835: v = a->a + a->i[i];
1836: sum = b[i];
1837: PetscSparseDenseMinusDot(sum,x,v,idx,n);
1838: t[i] = sum;
1839: x[i] = sum*idiag[i];
1840: }
1841: xb = t;
1842: PetscLogFlops(a->nz);
1843: } else xb = b;
1844: if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) {
1845: for (i=m-1; i>=0; i--) {
1846: n = a->i[i+1] - diag[i] - 1;
1847: idx = a->j + diag[i] + 1;
1848: v = a->a + diag[i] + 1;
1849: sum = xb[i];
1850: PetscSparseDenseMinusDot(sum,x,v,idx,n);
1851: if (xb == b) {
1852: x[i] = sum*idiag[i];
1853: } else {
1854: x[i] = (1-omega)*x[i] + sum*idiag[i]; /* omega in idiag */
1855: }
1856: }
1857: PetscLogFlops(a->nz); /* assumes 1/2 in upper */
1858: }
1859: its--;
1860: }
1861: while (its--) {
1862: if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) {
1863: for (i=0; i<m; i++) {
1864: /* lower */
1865: n = diag[i] - a->i[i];
1866: idx = a->j + a->i[i];
1867: v = a->a + a->i[i];
1868: sum = b[i];
1869: PetscSparseDenseMinusDot(sum,x,v,idx,n);
1870: t[i] = sum; /* save application of the lower-triangular part */
1871: /* upper */
1872: n = a->i[i+1] - diag[i] - 1;
1873: idx = a->j + diag[i] + 1;
1874: v = a->a + diag[i] + 1;
1875: PetscSparseDenseMinusDot(sum,x,v,idx,n);
1876: x[i] = (1. - omega)*x[i] + sum*idiag[i]; /* omega in idiag */
1877: }
1878: xb = t;
1879: PetscLogFlops(2.0*a->nz);
1880: } else xb = b;
1881: if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) {
1882: for (i=m-1; i>=0; i--) {
1883: sum = xb[i];
1884: if (xb == b) {
1885: /* whole matrix (no checkpointing available) */
1886: n = a->i[i+1] - a->i[i];
1887: idx = a->j + a->i[i];
1888: v = a->a + a->i[i];
1889: PetscSparseDenseMinusDot(sum,x,v,idx,n);
1890: x[i] = (1. - omega)*x[i] + (sum + mdiag[i]*x[i])*idiag[i];
1891: } else { /* lower-triangular part has been saved, so only apply upper-triangular */
1892: n = a->i[i+1] - diag[i] - 1;
1893: idx = a->j + diag[i] + 1;
1894: v = a->a + diag[i] + 1;
1895: PetscSparseDenseMinusDot(sum,x,v,idx,n);
1896: x[i] = (1. - omega)*x[i] + sum*idiag[i]; /* omega in idiag */
1897: }
1898: }
1899: if (xb == b) {
1900: PetscLogFlops(2.0*a->nz);
1901: } else {
1902: PetscLogFlops(a->nz); /* assumes 1/2 in upper */
1903: }
1904: }
1905: }
1906: VecRestoreArray(xx,&x);
1907: VecRestoreArrayRead(bb,&b);
1908: return(0);
1909: }
1914: PetscErrorCode MatGetInfo_SeqAIJ(Mat A,MatInfoType flag,MatInfo *info)
1915: {
1916: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1919: info->block_size = 1.0;
1920: info->nz_allocated = (double)a->maxnz;
1921: info->nz_used = (double)a->nz;
1922: info->nz_unneeded = (double)(a->maxnz - a->nz);
1923: info->assemblies = (double)A->num_ass;
1924: info->mallocs = (double)A->info.mallocs;
1925: info->memory = ((PetscObject)A)->mem;
1926: if (A->factortype) {
1927: info->fill_ratio_given = A->info.fill_ratio_given;
1928: info->fill_ratio_needed = A->info.fill_ratio_needed;
1929: info->factor_mallocs = A->info.factor_mallocs;
1930: } else {
1931: info->fill_ratio_given = 0;
1932: info->fill_ratio_needed = 0;
1933: info->factor_mallocs = 0;
1934: }
1935: return(0);
1936: }
1940: PetscErrorCode MatZeroRows_SeqAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b)
1941: {
1942: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1943: PetscInt i,m = A->rmap->n - 1,d = 0;
1944: PetscErrorCode ierr;
1945: const PetscScalar *xx;
1946: PetscScalar *bb;
1947: PetscBool missing;
1950: if (x && b) {
1951: VecGetArrayRead(x,&xx);
1952: VecGetArray(b,&bb);
1953: for (i=0; i<N; i++) {
1954: if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]);
1955: bb[rows[i]] = diag*xx[rows[i]];
1956: }
1957: VecRestoreArrayRead(x,&xx);
1958: VecRestoreArray(b,&bb);
1959: }
1961: if (a->keepnonzeropattern) {
1962: for (i=0; i<N; i++) {
1963: if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]);
1964: PetscMemzero(&a->a[a->i[rows[i]]],a->ilen[rows[i]]*sizeof(PetscScalar));
1965: }
1966: if (diag != 0.0) {
1967: MatMissingDiagonal_SeqAIJ(A,&missing,&d);
1968: if (missing) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry in row %D",d);
1969: for (i=0; i<N; i++) {
1970: a->a[a->diag[rows[i]]] = diag;
1971: }
1972: }
1973: } else {
1974: if (diag != 0.0) {
1975: for (i=0; i<N; i++) {
1976: if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]);
1977: if (a->ilen[rows[i]] > 0) {
1978: a->ilen[rows[i]] = 1;
1979: a->a[a->i[rows[i]]] = diag;
1980: a->j[a->i[rows[i]]] = rows[i];
1981: } else { /* in case row was completely empty */
1982: MatSetValues_SeqAIJ(A,1,&rows[i],1,&rows[i],&diag,INSERT_VALUES);
1983: }
1984: }
1985: } else {
1986: for (i=0; i<N; i++) {
1987: if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]);
1988: a->ilen[rows[i]] = 0;
1989: }
1990: }
1991: A->nonzerostate++;
1992: }
1993: MatAssemblyEnd_SeqAIJ(A,MAT_FINAL_ASSEMBLY);
1994: return(0);
1995: }
1999: PetscErrorCode MatZeroRowsColumns_SeqAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag,Vec x,Vec b)
2000: {
2001: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2002: PetscInt i,j,m = A->rmap->n - 1,d = 0;
2003: PetscErrorCode ierr;
2004: PetscBool missing,*zeroed,vecs = PETSC_FALSE;
2005: const PetscScalar *xx;
2006: PetscScalar *bb;
2009: if (x && b) {
2010: VecGetArrayRead(x,&xx);
2011: VecGetArray(b,&bb);
2012: vecs = PETSC_TRUE;
2013: }
2014: PetscCalloc1(A->rmap->n,&zeroed);
2015: for (i=0; i<N; i++) {
2016: if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]);
2017: PetscMemzero(&a->a[a->i[rows[i]]],a->ilen[rows[i]]*sizeof(PetscScalar));
2019: zeroed[rows[i]] = PETSC_TRUE;
2020: }
2021: for (i=0; i<A->rmap->n; i++) {
2022: if (!zeroed[i]) {
2023: for (j=a->i[i]; j<a->i[i+1]; j++) {
2024: if (zeroed[a->j[j]]) {
2025: if (vecs) bb[i] -= a->a[j]*xx[a->j[j]];
2026: a->a[j] = 0.0;
2027: }
2028: }
2029: } else if (vecs) bb[i] = diag*xx[i];
2030: }
2031: if (x && b) {
2032: VecRestoreArrayRead(x,&xx);
2033: VecRestoreArray(b,&bb);
2034: }
2035: PetscFree(zeroed);
2036: if (diag != 0.0) {
2037: MatMissingDiagonal_SeqAIJ(A,&missing,&d);
2038: if (missing) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry in row %D",d);
2039: for (i=0; i<N; i++) {
2040: a->a[a->diag[rows[i]]] = diag;
2041: }
2042: }
2043: MatAssemblyEnd_SeqAIJ(A,MAT_FINAL_ASSEMBLY);
2044: return(0);
2045: }
2049: PetscErrorCode MatGetRow_SeqAIJ(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v)
2050: {
2051: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2052: PetscInt *itmp;
2055: if (row < 0 || row >= A->rmap->n) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row %D out of range",row);
2057: *nz = a->i[row+1] - a->i[row];
2058: if (v) *v = a->a + a->i[row];
2059: if (idx) {
2060: itmp = a->j + a->i[row];
2061: if (*nz) *idx = itmp;
2062: else *idx = 0;
2063: }
2064: return(0);
2065: }
2067: /* remove this function? */
2070: PetscErrorCode MatRestoreRow_SeqAIJ(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v)
2071: {
2073: return(0);
2074: }
2078: PetscErrorCode MatNorm_SeqAIJ(Mat A,NormType type,PetscReal *nrm)
2079: {
2080: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2081: MatScalar *v = a->a;
2082: PetscReal sum = 0.0;
2084: PetscInt i,j;
2087: if (type == NORM_FROBENIUS) {
2088: for (i=0; i<a->nz; i++) {
2089: sum += PetscRealPart(PetscConj(*v)*(*v)); v++;
2090: }
2091: *nrm = PetscSqrtReal(sum);
2092: } else if (type == NORM_1) {
2093: PetscReal *tmp;
2094: PetscInt *jj = a->j;
2095: PetscCalloc1(A->cmap->n+1,&tmp);
2096: *nrm = 0.0;
2097: for (j=0; j<a->nz; j++) {
2098: tmp[*jj++] += PetscAbsScalar(*v); v++;
2099: }
2100: for (j=0; j<A->cmap->n; j++) {
2101: if (tmp[j] > *nrm) *nrm = tmp[j];
2102: }
2103: PetscFree(tmp);
2104: } else if (type == NORM_INFINITY) {
2105: *nrm = 0.0;
2106: for (j=0; j<A->rmap->n; j++) {
2107: v = a->a + a->i[j];
2108: sum = 0.0;
2109: for (i=0; i<a->i[j+1]-a->i[j]; i++) {
2110: sum += PetscAbsScalar(*v); v++;
2111: }
2112: if (sum > *nrm) *nrm = sum;
2113: }
2114: } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"No support for two norm");
2115: return(0);
2116: }
2118: /* Merged from MatGetSymbolicTranspose_SeqAIJ() - replace MatGetSymbolicTranspose_SeqAIJ()? */
2121: PetscErrorCode MatTransposeSymbolic_SeqAIJ(Mat A,Mat *B)
2122: {
2124: PetscInt i,j,anzj;
2125: Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data,*b;
2126: PetscInt an=A->cmap->N,am=A->rmap->N;
2127: PetscInt *ati,*atj,*atfill,*ai=a->i,*aj=a->j;
2130: /* Allocate space for symbolic transpose info and work array */
2131: PetscCalloc1((an+1),&ati);
2132: PetscMalloc1(ai[am],&atj);
2133: PetscMalloc1(an,&atfill);
2135: /* Walk through aj and count ## of non-zeros in each row of A^T. */
2136: /* Note: offset by 1 for fast conversion into csr format. */
2137: for (i=0;i<ai[am];i++) ati[aj[i]+1] += 1;
2138: /* Form ati for csr format of A^T. */
2139: for (i=0;i<an;i++) ati[i+1] += ati[i];
2141: /* Copy ati into atfill so we have locations of the next free space in atj */
2142: PetscMemcpy(atfill,ati,an*sizeof(PetscInt));
2144: /* Walk through A row-wise and mark nonzero entries of A^T. */
2145: for (i=0;i<am;i++) {
2146: anzj = ai[i+1] - ai[i];
2147: for (j=0;j<anzj;j++) {
2148: atj[atfill[*aj]] = i;
2149: atfill[*aj++] += 1;
2150: }
2151: }
2153: /* Clean up temporary space and complete requests. */
2154: PetscFree(atfill);
2155: MatCreateSeqAIJWithArrays(PetscObjectComm((PetscObject)A),an,am,ati,atj,NULL,B);
2156: MatSetBlockSizes(*B,PetscAbs(A->cmap->bs),PetscAbs(A->rmap->bs));
2158: b = (Mat_SeqAIJ*)((*B)->data);
2159: b->free_a = PETSC_FALSE;
2160: b->free_ij = PETSC_TRUE;
2161: b->nonew = 0;
2162: return(0);
2163: }
2167: PetscErrorCode MatTranspose_SeqAIJ(Mat A,MatReuse reuse,Mat *B)
2168: {
2169: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2170: Mat C;
2172: PetscInt i,*aj = a->j,*ai = a->i,m = A->rmap->n,len,*col;
2173: MatScalar *array = a->a;
2176: if (reuse == MAT_REUSE_MATRIX && A == *B && m != A->cmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Square matrix only for in-place");
2178: if (reuse == MAT_INITIAL_MATRIX || *B == A) {
2179: PetscCalloc1((1+A->cmap->n),&col);
2181: for (i=0; i<ai[m]; i++) col[aj[i]] += 1;
2182: MatCreate(PetscObjectComm((PetscObject)A),&C);
2183: MatSetSizes(C,A->cmap->n,m,A->cmap->n,m);
2184: MatSetBlockSizes(C,PetscAbs(A->cmap->bs),PetscAbs(A->rmap->bs));
2185: MatSetType(C,((PetscObject)A)->type_name);
2186: MatSeqAIJSetPreallocation_SeqAIJ(C,0,col);
2187: PetscFree(col);
2188: } else {
2189: C = *B;
2190: }
2192: for (i=0; i<m; i++) {
2193: len = ai[i+1]-ai[i];
2194: MatSetValues_SeqAIJ(C,len,aj,1,&i,array,INSERT_VALUES);
2195: array += len;
2196: aj += len;
2197: }
2198: MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);
2199: MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);
2201: if (reuse == MAT_INITIAL_MATRIX || *B != A) {
2202: *B = C;
2203: } else {
2204: MatHeaderMerge(A,C);
2205: }
2206: return(0);
2207: }
2211: PetscErrorCode MatIsTranspose_SeqAIJ(Mat A,Mat B,PetscReal tol,PetscBool *f)
2212: {
2213: Mat_SeqAIJ *aij = (Mat_SeqAIJ*) A->data,*bij = (Mat_SeqAIJ*) A->data;
2214: PetscInt *adx,*bdx,*aii,*bii,*aptr,*bptr;
2215: MatScalar *va,*vb;
2217: PetscInt ma,na,mb,nb, i;
2220: bij = (Mat_SeqAIJ*) B->data;
2222: MatGetSize(A,&ma,&na);
2223: MatGetSize(B,&mb,&nb);
2224: if (ma!=nb || na!=mb) {
2225: *f = PETSC_FALSE;
2226: return(0);
2227: }
2228: aii = aij->i; bii = bij->i;
2229: adx = aij->j; bdx = bij->j;
2230: va = aij->a; vb = bij->a;
2231: PetscMalloc1(ma,&aptr);
2232: PetscMalloc1(mb,&bptr);
2233: for (i=0; i<ma; i++) aptr[i] = aii[i];
2234: for (i=0; i<mb; i++) bptr[i] = bii[i];
2236: *f = PETSC_TRUE;
2237: for (i=0; i<ma; i++) {
2238: while (aptr[i]<aii[i+1]) {
2239: PetscInt idc,idr;
2240: PetscScalar vc,vr;
2241: /* column/row index/value */
2242: idc = adx[aptr[i]];
2243: idr = bdx[bptr[idc]];
2244: vc = va[aptr[i]];
2245: vr = vb[bptr[idc]];
2246: if (i!=idr || PetscAbsScalar(vc-vr) > tol) {
2247: *f = PETSC_FALSE;
2248: goto done;
2249: } else {
2250: aptr[i]++;
2251: if (B || i!=idc) bptr[idc]++;
2252: }
2253: }
2254: }
2255: done:
2256: PetscFree(aptr);
2257: PetscFree(bptr);
2258: return(0);
2259: }
2263: PetscErrorCode MatIsHermitianTranspose_SeqAIJ(Mat A,Mat B,PetscReal tol,PetscBool *f)
2264: {
2265: Mat_SeqAIJ *aij = (Mat_SeqAIJ*) A->data,*bij = (Mat_SeqAIJ*) A->data;
2266: PetscInt *adx,*bdx,*aii,*bii,*aptr,*bptr;
2267: MatScalar *va,*vb;
2269: PetscInt ma,na,mb,nb, i;
2272: bij = (Mat_SeqAIJ*) B->data;
2274: MatGetSize(A,&ma,&na);
2275: MatGetSize(B,&mb,&nb);
2276: if (ma!=nb || na!=mb) {
2277: *f = PETSC_FALSE;
2278: return(0);
2279: }
2280: aii = aij->i; bii = bij->i;
2281: adx = aij->j; bdx = bij->j;
2282: va = aij->a; vb = bij->a;
2283: PetscMalloc1(ma,&aptr);
2284: PetscMalloc1(mb,&bptr);
2285: for (i=0; i<ma; i++) aptr[i] = aii[i];
2286: for (i=0; i<mb; i++) bptr[i] = bii[i];
2288: *f = PETSC_TRUE;
2289: for (i=0; i<ma; i++) {
2290: while (aptr[i]<aii[i+1]) {
2291: PetscInt idc,idr;
2292: PetscScalar vc,vr;
2293: /* column/row index/value */
2294: idc = adx[aptr[i]];
2295: idr = bdx[bptr[idc]];
2296: vc = va[aptr[i]];
2297: vr = vb[bptr[idc]];
2298: if (i!=idr || PetscAbsScalar(vc-PetscConj(vr)) > tol) {
2299: *f = PETSC_FALSE;
2300: goto done;
2301: } else {
2302: aptr[i]++;
2303: if (B || i!=idc) bptr[idc]++;
2304: }
2305: }
2306: }
2307: done:
2308: PetscFree(aptr);
2309: PetscFree(bptr);
2310: return(0);
2311: }
2315: PetscErrorCode MatIsSymmetric_SeqAIJ(Mat A,PetscReal tol,PetscBool *f)
2316: {
2320: MatIsTranspose_SeqAIJ(A,A,tol,f);
2321: return(0);
2322: }
2326: PetscErrorCode MatIsHermitian_SeqAIJ(Mat A,PetscReal tol,PetscBool *f)
2327: {
2331: MatIsHermitianTranspose_SeqAIJ(A,A,tol,f);
2332: return(0);
2333: }
2337: PetscErrorCode MatDiagonalScale_SeqAIJ(Mat A,Vec ll,Vec rr)
2338: {
2339: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2340: PetscScalar *l,*r,x;
2341: MatScalar *v;
2343: PetscInt i,j,m = A->rmap->n,n = A->cmap->n,M,nz = a->nz,*jj;
2346: if (ll) {
2347: /* The local size is used so that VecMPI can be passed to this routine
2348: by MatDiagonalScale_MPIAIJ */
2349: VecGetLocalSize(ll,&m);
2350: if (m != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Left scaling vector wrong length");
2351: VecGetArray(ll,&l);
2352: v = a->a;
2353: for (i=0; i<m; i++) {
2354: x = l[i];
2355: M = a->i[i+1] - a->i[i];
2356: for (j=0; j<M; j++) (*v++) *= x;
2357: }
2358: VecRestoreArray(ll,&l);
2359: PetscLogFlops(nz);
2360: }
2361: if (rr) {
2362: VecGetLocalSize(rr,&n);
2363: if (n != A->cmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Right scaling vector wrong length");
2364: VecGetArray(rr,&r);
2365: v = a->a; jj = a->j;
2366: for (i=0; i<nz; i++) (*v++) *= r[*jj++];
2367: VecRestoreArray(rr,&r);
2368: PetscLogFlops(nz);
2369: }
2370: MatSeqAIJInvalidateDiagonal(A);
2371: return(0);
2372: }
2376: PetscErrorCode MatGetSubMatrix_SeqAIJ(Mat A,IS isrow,IS iscol,PetscInt csize,MatReuse scall,Mat *B)
2377: {
2378: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*c;
2380: PetscInt *smap,i,k,kstart,kend,oldcols = A->cmap->n,*lens;
2381: PetscInt row,mat_i,*mat_j,tcol,first,step,*mat_ilen,sum,lensi;
2382: const PetscInt *irow,*icol;
2383: PetscInt nrows,ncols;
2384: PetscInt *starts,*j_new,*i_new,*aj = a->j,*ai = a->i,ii,*ailen = a->ilen;
2385: MatScalar *a_new,*mat_a;
2386: Mat C;
2387: PetscBool stride,sorted;
2390: ISSorted(isrow,&sorted);
2391: if (!sorted) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"ISrow is not sorted");
2392: ISSorted(iscol,&sorted);
2393: if (!sorted) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"IScol is not sorted");
2395: ISGetIndices(isrow,&irow);
2396: ISGetLocalSize(isrow,&nrows);
2397: ISGetLocalSize(iscol,&ncols);
2399: PetscObjectTypeCompare((PetscObject)iscol,ISSTRIDE,&stride);
2400: if (stride) {
2401: ISStrideGetInfo(iscol,&first,&step);
2402: } else {
2403: first = 0;
2404: step = 0;
2405: }
2406: if (stride && step == 1) {
2407: /* special case of contiguous rows */
2408: PetscMalloc2(nrows,&lens,nrows,&starts);
2409: /* loop over new rows determining lens and starting points */
2410: for (i=0; i<nrows; i++) {
2411: kstart = ai[irow[i]];
2412: kend = kstart + ailen[irow[i]];
2413: for (k=kstart; k<kend; k++) {
2414: if (aj[k] >= first) {
2415: starts[i] = k;
2416: break;
2417: }
2418: }
2419: sum = 0;
2420: while (k < kend) {
2421: if (aj[k++] >= first+ncols) break;
2422: sum++;
2423: }
2424: lens[i] = sum;
2425: }
2426: /* create submatrix */
2427: if (scall == MAT_REUSE_MATRIX) {
2428: PetscInt n_cols,n_rows;
2429: MatGetSize(*B,&n_rows,&n_cols);
2430: if (n_rows != nrows || n_cols != ncols) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Reused submatrix wrong size");
2431: MatZeroEntries(*B);
2432: C = *B;
2433: } else {
2434: PetscInt rbs,cbs;
2435: MatCreate(PetscObjectComm((PetscObject)A),&C);
2436: MatSetSizes(C,nrows,ncols,PETSC_DETERMINE,PETSC_DETERMINE);
2437: ISGetBlockSize(isrow,&rbs);
2438: ISGetBlockSize(iscol,&cbs);
2439: MatSetBlockSizes(C,rbs,cbs);
2440: MatSetType(C,((PetscObject)A)->type_name);
2441: MatSeqAIJSetPreallocation_SeqAIJ(C,0,lens);
2442: }
2443: c = (Mat_SeqAIJ*)C->data;
2445: /* loop over rows inserting into submatrix */
2446: a_new = c->a;
2447: j_new = c->j;
2448: i_new = c->i;
2450: for (i=0; i<nrows; i++) {
2451: ii = starts[i];
2452: lensi = lens[i];
2453: for (k=0; k<lensi; k++) {
2454: *j_new++ = aj[ii+k] - first;
2455: }
2456: PetscMemcpy(a_new,a->a + starts[i],lensi*sizeof(PetscScalar));
2457: a_new += lensi;
2458: i_new[i+1] = i_new[i] + lensi;
2459: c->ilen[i] = lensi;
2460: }
2461: PetscFree2(lens,starts);
2462: } else {
2463: ISGetIndices(iscol,&icol);
2464: PetscCalloc1(oldcols,&smap);
2465: PetscMalloc1((1+nrows),&lens);
2466: for (i=0; i<ncols; i++) {
2467: #if defined(PETSC_USE_DEBUG)
2468: if (icol[i] >= oldcols) SETERRQ3(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Requesting column beyond largest column icol[%D] %D <= A->cmap->n %D",i,icol[i],oldcols);
2469: #endif
2470: smap[icol[i]] = i+1;
2471: }
2473: /* determine lens of each row */
2474: for (i=0; i<nrows; i++) {
2475: kstart = ai[irow[i]];
2476: kend = kstart + a->ilen[irow[i]];
2477: lens[i] = 0;
2478: for (k=kstart; k<kend; k++) {
2479: if (smap[aj[k]]) {
2480: lens[i]++;
2481: }
2482: }
2483: }
2484: /* Create and fill new matrix */
2485: if (scall == MAT_REUSE_MATRIX) {
2486: PetscBool equal;
2488: c = (Mat_SeqAIJ*)((*B)->data);
2489: if ((*B)->rmap->n != nrows || (*B)->cmap->n != ncols) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. wrong size");
2490: PetscMemcmp(c->ilen,lens,(*B)->rmap->n*sizeof(PetscInt),&equal);
2491: if (!equal) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. wrong no of nonzeros");
2492: PetscMemzero(c->ilen,(*B)->rmap->n*sizeof(PetscInt));
2493: C = *B;
2494: } else {
2495: PetscInt rbs,cbs;
2496: MatCreate(PetscObjectComm((PetscObject)A),&C);
2497: MatSetSizes(C,nrows,ncols,PETSC_DETERMINE,PETSC_DETERMINE);
2498: ISGetBlockSize(isrow,&rbs);
2499: ISGetBlockSize(iscol,&cbs);
2500: MatSetBlockSizes(C,rbs,cbs);
2501: MatSetType(C,((PetscObject)A)->type_name);
2502: MatSeqAIJSetPreallocation_SeqAIJ(C,0,lens);
2503: }
2504: c = (Mat_SeqAIJ*)(C->data);
2505: for (i=0; i<nrows; i++) {
2506: row = irow[i];
2507: kstart = ai[row];
2508: kend = kstart + a->ilen[row];
2509: mat_i = c->i[i];
2510: mat_j = c->j + mat_i;
2511: mat_a = c->a + mat_i;
2512: mat_ilen = c->ilen + i;
2513: for (k=kstart; k<kend; k++) {
2514: if ((tcol=smap[a->j[k]])) {
2515: *mat_j++ = tcol - 1;
2516: *mat_a++ = a->a[k];
2517: (*mat_ilen)++;
2519: }
2520: }
2521: }
2522: /* Free work space */
2523: ISRestoreIndices(iscol,&icol);
2524: PetscFree(smap);
2525: PetscFree(lens);
2526: }
2527: MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);
2528: MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);
2530: ISRestoreIndices(isrow,&irow);
2531: *B = C;
2532: return(0);
2533: }
2537: PetscErrorCode MatGetMultiProcBlock_SeqAIJ(Mat mat,MPI_Comm subComm,MatReuse scall,Mat *subMat)
2538: {
2540: Mat B;
2543: if (scall == MAT_INITIAL_MATRIX) {
2544: MatCreate(subComm,&B);
2545: MatSetSizes(B,mat->rmap->n,mat->cmap->n,mat->rmap->n,mat->cmap->n);
2546: MatSetBlockSizesFromMats(B,mat,mat);
2547: MatSetType(B,MATSEQAIJ);
2548: MatDuplicateNoCreate_SeqAIJ(B,mat,MAT_COPY_VALUES,PETSC_TRUE);
2549: *subMat = B;
2550: } else {
2551: MatCopy_SeqAIJ(mat,*subMat,SAME_NONZERO_PATTERN);
2552: }
2553: return(0);
2554: }
2558: PetscErrorCode MatILUFactor_SeqAIJ(Mat inA,IS row,IS col,const MatFactorInfo *info)
2559: {
2560: Mat_SeqAIJ *a = (Mat_SeqAIJ*)inA->data;
2562: Mat outA;
2563: PetscBool row_identity,col_identity;
2566: if (info->levels != 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Only levels=0 supported for in-place ilu");
2568: ISIdentity(row,&row_identity);
2569: ISIdentity(col,&col_identity);
2571: outA = inA;
2572: outA->factortype = MAT_FACTOR_LU;
2574: PetscObjectReference((PetscObject)row);
2575: ISDestroy(&a->row);
2577: a->row = row;
2579: PetscObjectReference((PetscObject)col);
2580: ISDestroy(&a->col);
2582: a->col = col;
2584: /* Create the inverse permutation so that it can be used in MatLUFactorNumeric() */
2585: ISDestroy(&a->icol);
2586: ISInvertPermutation(col,PETSC_DECIDE,&a->icol);
2587: PetscLogObjectParent((PetscObject)inA,(PetscObject)a->icol);
2589: if (!a->solve_work) { /* this matrix may have been factored before */
2590: PetscMalloc1((inA->rmap->n+1),&a->solve_work);
2591: PetscLogObjectMemory((PetscObject)inA, (inA->rmap->n+1)*sizeof(PetscScalar));
2592: }
2594: MatMarkDiagonal_SeqAIJ(inA);
2595: if (row_identity && col_identity) {
2596: MatLUFactorNumeric_SeqAIJ_inplace(outA,inA,info);
2597: } else {
2598: MatLUFactorNumeric_SeqAIJ_InplaceWithPerm(outA,inA,info);
2599: }
2600: return(0);
2601: }
2605: PetscErrorCode MatScale_SeqAIJ(Mat inA,PetscScalar alpha)
2606: {
2607: Mat_SeqAIJ *a = (Mat_SeqAIJ*)inA->data;
2608: PetscScalar oalpha = alpha;
2610: PetscBLASInt one = 1,bnz;
2613: PetscBLASIntCast(a->nz,&bnz);
2614: PetscStackCallBLAS("BLASscal",BLASscal_(&bnz,&oalpha,a->a,&one));
2615: PetscLogFlops(a->nz);
2616: MatSeqAIJInvalidateDiagonal(inA);
2617: return(0);
2618: }
2622: PetscErrorCode MatGetSubMatrices_SeqAIJ(Mat A,PetscInt n,const IS irow[],const IS icol[],MatReuse scall,Mat *B[])
2623: {
2625: PetscInt i;
2628: if (scall == MAT_INITIAL_MATRIX) {
2629: PetscMalloc1((n+1),B);
2630: }
2632: for (i=0; i<n; i++) {
2633: MatGetSubMatrix_SeqAIJ(A,irow[i],icol[i],PETSC_DECIDE,scall,&(*B)[i]);
2634: }
2635: return(0);
2636: }
2640: PetscErrorCode MatIncreaseOverlap_SeqAIJ(Mat A,PetscInt is_max,IS is[],PetscInt ov)
2641: {
2642: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2644: PetscInt row,i,j,k,l,m,n,*nidx,isz,val;
2645: const PetscInt *idx;
2646: PetscInt start,end,*ai,*aj;
2647: PetscBT table;
2650: m = A->rmap->n;
2651: ai = a->i;
2652: aj = a->j;
2654: if (ov < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"illegal negative overlap value used");
2656: PetscMalloc1((m+1),&nidx);
2657: PetscBTCreate(m,&table);
2659: for (i=0; i<is_max; i++) {
2660: /* Initialize the two local arrays */
2661: isz = 0;
2662: PetscBTMemzero(m,table);
2664: /* Extract the indices, assume there can be duplicate entries */
2665: ISGetIndices(is[i],&idx);
2666: ISGetLocalSize(is[i],&n);
2668: /* Enter these into the temp arrays. I.e., mark table[row], enter row into new index */
2669: for (j=0; j<n; ++j) {
2670: if (!PetscBTLookupSet(table,idx[j])) nidx[isz++] = idx[j];
2671: }
2672: ISRestoreIndices(is[i],&idx);
2673: ISDestroy(&is[i]);
2675: k = 0;
2676: for (j=0; j<ov; j++) { /* for each overlap */
2677: n = isz;
2678: for (; k<n; k++) { /* do only those rows in nidx[k], which are not done yet */
2679: row = nidx[k];
2680: start = ai[row];
2681: end = ai[row+1];
2682: for (l = start; l<end; l++) {
2683: val = aj[l];
2684: if (!PetscBTLookupSet(table,val)) nidx[isz++] = val;
2685: }
2686: }
2687: }
2688: ISCreateGeneral(PETSC_COMM_SELF,isz,nidx,PETSC_COPY_VALUES,(is+i));
2689: }
2690: PetscBTDestroy(&table);
2691: PetscFree(nidx);
2692: return(0);
2693: }
2695: /* -------------------------------------------------------------- */
2698: PetscErrorCode MatPermute_SeqAIJ(Mat A,IS rowp,IS colp,Mat *B)
2699: {
2700: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2702: PetscInt i,nz = 0,m = A->rmap->n,n = A->cmap->n;
2703: const PetscInt *row,*col;
2704: PetscInt *cnew,j,*lens;
2705: IS icolp,irowp;
2706: PetscInt *cwork = NULL;
2707: PetscScalar *vwork = NULL;
2710: ISInvertPermutation(rowp,PETSC_DECIDE,&irowp);
2711: ISGetIndices(irowp,&row);
2712: ISInvertPermutation(colp,PETSC_DECIDE,&icolp);
2713: ISGetIndices(icolp,&col);
2715: /* determine lengths of permuted rows */
2716: PetscMalloc1((m+1),&lens);
2717: for (i=0; i<m; i++) lens[row[i]] = a->i[i+1] - a->i[i];
2718: MatCreate(PetscObjectComm((PetscObject)A),B);
2719: MatSetSizes(*B,m,n,m,n);
2720: MatSetBlockSizesFromMats(*B,A,A);
2721: MatSetType(*B,((PetscObject)A)->type_name);
2722: MatSeqAIJSetPreallocation_SeqAIJ(*B,0,lens);
2723: PetscFree(lens);
2725: PetscMalloc1(n,&cnew);
2726: for (i=0; i<m; i++) {
2727: MatGetRow_SeqAIJ(A,i,&nz,&cwork,&vwork);
2728: for (j=0; j<nz; j++) cnew[j] = col[cwork[j]];
2729: MatSetValues_SeqAIJ(*B,1,&row[i],nz,cnew,vwork,INSERT_VALUES);
2730: MatRestoreRow_SeqAIJ(A,i,&nz,&cwork,&vwork);
2731: }
2732: PetscFree(cnew);
2734: (*B)->assembled = PETSC_FALSE;
2736: MatAssemblyBegin(*B,MAT_FINAL_ASSEMBLY);
2737: MatAssemblyEnd(*B,MAT_FINAL_ASSEMBLY);
2738: ISRestoreIndices(irowp,&row);
2739: ISRestoreIndices(icolp,&col);
2740: ISDestroy(&irowp);
2741: ISDestroy(&icolp);
2742: return(0);
2743: }
2747: PetscErrorCode MatCopy_SeqAIJ(Mat A,Mat B,MatStructure str)
2748: {
2752: /* If the two matrices have the same copy implementation, use fast copy. */
2753: if (str == SAME_NONZERO_PATTERN && (A->ops->copy == B->ops->copy)) {
2754: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2755: Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data;
2757: if (a->i[A->rmap->n] != b->i[B->rmap->n]) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"Number of nonzeros in two matrices are different");
2758: PetscMemcpy(b->a,a->a,(a->i[A->rmap->n])*sizeof(PetscScalar));
2759: } else {
2760: MatCopy_Basic(A,B,str);
2761: }
2762: return(0);
2763: }
2767: PetscErrorCode MatSetUp_SeqAIJ(Mat A)
2768: {
2772: MatSeqAIJSetPreallocation_SeqAIJ(A,PETSC_DEFAULT,0);
2773: return(0);
2774: }
2778: PetscErrorCode MatSeqAIJGetArray_SeqAIJ(Mat A,PetscScalar *array[])
2779: {
2780: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2783: *array = a->a;
2784: return(0);
2785: }
2789: PetscErrorCode MatSeqAIJRestoreArray_SeqAIJ(Mat A,PetscScalar *array[])
2790: {
2792: return(0);
2793: }
2795: /*
2796: Computes the number of nonzeros per row needed for preallocation when X and Y
2797: have different nonzero structure.
2798: */
2801: PetscErrorCode MatAXPYGetPreallocation_SeqAIJ(Mat Y,Mat X,PetscInt *nnz)
2802: {
2803: PetscInt i,m=Y->rmap->N;
2804: Mat_SeqAIJ *x = (Mat_SeqAIJ*)X->data;
2805: Mat_SeqAIJ *y = (Mat_SeqAIJ*)Y->data;
2806: const PetscInt *xi = x->i,*yi = y->i;
2809: /* Set the number of nonzeros in the new matrix */
2810: for (i=0; i<m; i++) {
2811: PetscInt j,k,nzx = xi[i+1] - xi[i],nzy = yi[i+1] - yi[i];
2812: const PetscInt *xj = x->j+xi[i],*yj = y->j+yi[i];
2813: nnz[i] = 0;
2814: for (j=0,k=0; j<nzx; j++) { /* Point in X */
2815: for (; k<nzy && yj[k]<xj[j]; k++) nnz[i]++; /* Catch up to X */
2816: if (k<nzy && yj[k]==xj[j]) k++; /* Skip duplicate */
2817: nnz[i]++;
2818: }
2819: for (; k<nzy; k++) nnz[i]++;
2820: }
2821: return(0);
2822: }
2826: PetscErrorCode MatAXPY_SeqAIJ(Mat Y,PetscScalar a,Mat X,MatStructure str)
2827: {
2829: PetscInt i;
2830: Mat_SeqAIJ *x = (Mat_SeqAIJ*)X->data,*y = (Mat_SeqAIJ*)Y->data;
2831: PetscBLASInt one=1,bnz;
2834: PetscBLASIntCast(x->nz,&bnz);
2835: if (str == SAME_NONZERO_PATTERN) {
2836: PetscScalar alpha = a;
2837: PetscStackCallBLAS("BLASaxpy",BLASaxpy_(&bnz,&alpha,x->a,&one,y->a,&one));
2838: MatSeqAIJInvalidateDiagonal(Y);
2839: PetscObjectStateIncrease((PetscObject)Y);
2840: } else if (str == SUBSET_NONZERO_PATTERN) { /* nonzeros of X is a subset of Y's */
2841: if (y->xtoy && y->XtoY != X) {
2842: PetscFree(y->xtoy);
2843: MatDestroy(&y->XtoY);
2844: }
2845: if (!y->xtoy) { /* get xtoy */
2846: MatAXPYGetxtoy_Private(X->rmap->n,x->i,x->j,NULL, y->i,y->j,NULL, &y->xtoy);
2847: y->XtoY = X;
2848: PetscObjectReference((PetscObject)X);
2849: }
2850: for (i=0; i<x->nz; i++) y->a[y->xtoy[i]] += a*(x->a[i]);
2851: PetscObjectStateIncrease((PetscObject)Y);
2852: PetscInfo3(Y,"ratio of nnz(X)/nnz(Y): %D/%D = %g\n",x->nz,y->nz,(double)((PetscReal)(x->nz)/(y->nz+1)));
2853: } else {
2854: Mat B;
2855: PetscInt *nnz;
2856: PetscMalloc1(Y->rmap->N,&nnz);
2857: MatCreate(PetscObjectComm((PetscObject)Y),&B);
2858: PetscObjectSetName((PetscObject)B,((PetscObject)Y)->name);
2859: MatSetSizes(B,Y->rmap->n,Y->cmap->n,Y->rmap->N,Y->cmap->N);
2860: MatSetBlockSizesFromMats(B,Y,Y);
2861: MatSetType(B,(MatType) ((PetscObject)Y)->type_name);
2862: MatAXPYGetPreallocation_SeqAIJ(Y,X,nnz);
2863: MatSeqAIJSetPreallocation(B,0,nnz);
2864: MatAXPY_BasicWithPreallocation(B,Y,a,X,str);
2865: MatHeaderReplace(Y,B);
2866: PetscFree(nnz);
2867: }
2868: return(0);
2869: }
2873: PetscErrorCode MatConjugate_SeqAIJ(Mat mat)
2874: {
2875: #if defined(PETSC_USE_COMPLEX)
2876: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data;
2877: PetscInt i,nz;
2878: PetscScalar *a;
2881: nz = aij->nz;
2882: a = aij->a;
2883: for (i=0; i<nz; i++) a[i] = PetscConj(a[i]);
2884: #else
2886: #endif
2887: return(0);
2888: }
2892: PetscErrorCode MatGetRowMaxAbs_SeqAIJ(Mat A,Vec v,PetscInt idx[])
2893: {
2894: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2896: PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n;
2897: PetscReal atmp;
2898: PetscScalar *x;
2899: MatScalar *aa;
2902: if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
2903: aa = a->a;
2904: ai = a->i;
2905: aj = a->j;
2907: VecSet(v,0.0);
2908: VecGetArray(v,&x);
2909: VecGetLocalSize(v,&n);
2910: if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector");
2911: for (i=0; i<m; i++) {
2912: ncols = ai[1] - ai[0]; ai++;
2913: x[i] = 0.0;
2914: for (j=0; j<ncols; j++) {
2915: atmp = PetscAbsScalar(*aa);
2916: if (PetscAbsScalar(x[i]) < atmp) {x[i] = atmp; if (idx) idx[i] = *aj;}
2917: aa++; aj++;
2918: }
2919: }
2920: VecRestoreArray(v,&x);
2921: return(0);
2922: }
2926: PetscErrorCode MatGetRowMax_SeqAIJ(Mat A,Vec v,PetscInt idx[])
2927: {
2928: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2930: PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n;
2931: PetscScalar *x;
2932: MatScalar *aa;
2935: if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
2936: aa = a->a;
2937: ai = a->i;
2938: aj = a->j;
2940: VecSet(v,0.0);
2941: VecGetArray(v,&x);
2942: VecGetLocalSize(v,&n);
2943: if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector");
2944: for (i=0; i<m; i++) {
2945: ncols = ai[1] - ai[0]; ai++;
2946: if (ncols == A->cmap->n) { /* row is dense */
2947: x[i] = *aa; if (idx) idx[i] = 0;
2948: } else { /* row is sparse so already KNOW maximum is 0.0 or higher */
2949: x[i] = 0.0;
2950: if (idx) {
2951: idx[i] = 0; /* in case ncols is zero */
2952: for (j=0;j<ncols;j++) { /* find first implicit 0.0 in the row */
2953: if (aj[j] > j) {
2954: idx[i] = j;
2955: break;
2956: }
2957: }
2958: }
2959: }
2960: for (j=0; j<ncols; j++) {
2961: if (PetscRealPart(x[i]) < PetscRealPart(*aa)) {x[i] = *aa; if (idx) idx[i] = *aj;}
2962: aa++; aj++;
2963: }
2964: }
2965: VecRestoreArray(v,&x);
2966: return(0);
2967: }
2971: PetscErrorCode MatGetRowMinAbs_SeqAIJ(Mat A,Vec v,PetscInt idx[])
2972: {
2973: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2975: PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n;
2976: PetscReal atmp;
2977: PetscScalar *x;
2978: MatScalar *aa;
2981: if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
2982: aa = a->a;
2983: ai = a->i;
2984: aj = a->j;
2986: VecSet(v,0.0);
2987: VecGetArray(v,&x);
2988: VecGetLocalSize(v,&n);
2989: if (n != A->rmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector, %D vs. %D rows", A->rmap->n, n);
2990: for (i=0; i<m; i++) {
2991: ncols = ai[1] - ai[0]; ai++;
2992: if (ncols) {
2993: /* Get first nonzero */
2994: for (j = 0; j < ncols; j++) {
2995: atmp = PetscAbsScalar(aa[j]);
2996: if (atmp > 1.0e-12) {
2997: x[i] = atmp;
2998: if (idx) idx[i] = aj[j];
2999: break;
3000: }
3001: }
3002: if (j == ncols) {x[i] = PetscAbsScalar(*aa); if (idx) idx[i] = *aj;}
3003: } else {
3004: x[i] = 0.0; if (idx) idx[i] = 0;
3005: }
3006: for (j = 0; j < ncols; j++) {
3007: atmp = PetscAbsScalar(*aa);
3008: if (atmp > 1.0e-12 && PetscAbsScalar(x[i]) > atmp) {x[i] = atmp; if (idx) idx[i] = *aj;}
3009: aa++; aj++;
3010: }
3011: }
3012: VecRestoreArray(v,&x);
3013: return(0);
3014: }
3018: PetscErrorCode MatGetRowMin_SeqAIJ(Mat A,Vec v,PetscInt idx[])
3019: {
3020: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
3022: PetscInt i,j,m = A->rmap->n,*ai,*aj,ncols,n;
3023: PetscScalar *x;
3024: MatScalar *aa;
3027: if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
3028: aa = a->a;
3029: ai = a->i;
3030: aj = a->j;
3032: VecSet(v,0.0);
3033: VecGetArray(v,&x);
3034: VecGetLocalSize(v,&n);
3035: if (n != A->rmap->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector");
3036: for (i=0; i<m; i++) {
3037: ncols = ai[1] - ai[0]; ai++;
3038: if (ncols == A->cmap->n) { /* row is dense */
3039: x[i] = *aa; if (idx) idx[i] = 0;
3040: } else { /* row is sparse so already KNOW minimum is 0.0 or lower */
3041: x[i] = 0.0;
3042: if (idx) { /* find first implicit 0.0 in the row */
3043: idx[i] = 0; /* in case ncols is zero */
3044: for (j=0; j<ncols; j++) {
3045: if (aj[j] > j) {
3046: idx[i] = j;
3047: break;
3048: }
3049: }
3050: }
3051: }
3052: for (j=0; j<ncols; j++) {
3053: if (PetscRealPart(x[i]) > PetscRealPart(*aa)) {x[i] = *aa; if (idx) idx[i] = *aj;}
3054: aa++; aj++;
3055: }
3056: }
3057: VecRestoreArray(v,&x);
3058: return(0);
3059: }
3061: #include <petscblaslapack.h>
3062: #include <petsc-private/kernels/blockinvert.h>
3066: PetscErrorCode MatInvertBlockDiagonal_SeqAIJ(Mat A,const PetscScalar **values)
3067: {
3068: Mat_SeqAIJ *a = (Mat_SeqAIJ*) A->data;
3070: PetscInt i,bs = PetscAbs(A->rmap->bs),mbs = A->rmap->n/bs,ipvt[5],bs2 = bs*bs,*v_pivots,ij[7],*IJ,j;
3071: MatScalar *diag,work[25],*v_work;
3072: PetscReal shift = 0.0;
3075: if (a->ibdiagvalid) {
3076: if (values) *values = a->ibdiag;
3077: return(0);
3078: }
3079: MatMarkDiagonal_SeqAIJ(A);
3080: if (!a->ibdiag) {
3081: PetscMalloc1(bs2*mbs,&a->ibdiag);
3082: PetscLogObjectMemory((PetscObject)A,bs2*mbs*sizeof(PetscScalar));
3083: }
3084: diag = a->ibdiag;
3085: if (values) *values = a->ibdiag;
3086: /* factor and invert each block */
3087: switch (bs) {
3088: case 1:
3089: for (i=0; i<mbs; i++) {
3090: MatGetValues(A,1,&i,1,&i,diag+i);
3091: diag[i] = (PetscScalar)1.0 / (diag[i] + shift);
3092: }
3093: break;
3094: case 2:
3095: for (i=0; i<mbs; i++) {
3096: ij[0] = 2*i; ij[1] = 2*i + 1;
3097: MatGetValues(A,2,ij,2,ij,diag);
3098: PetscKernel_A_gets_inverse_A_2(diag,shift);
3099: PetscKernel_A_gets_transpose_A_2(diag);
3100: diag += 4;
3101: }
3102: break;
3103: case 3:
3104: for (i=0; i<mbs; i++) {
3105: ij[0] = 3*i; ij[1] = 3*i + 1; ij[2] = 3*i + 2;
3106: MatGetValues(A,3,ij,3,ij,diag);
3107: PetscKernel_A_gets_inverse_A_3(diag,shift);
3108: PetscKernel_A_gets_transpose_A_3(diag);
3109: diag += 9;
3110: }
3111: break;
3112: case 4:
3113: for (i=0; i<mbs; i++) {
3114: ij[0] = 4*i; ij[1] = 4*i + 1; ij[2] = 4*i + 2; ij[3] = 4*i + 3;
3115: MatGetValues(A,4,ij,4,ij,diag);
3116: PetscKernel_A_gets_inverse_A_4(diag,shift);
3117: PetscKernel_A_gets_transpose_A_4(diag);
3118: diag += 16;
3119: }
3120: break;
3121: case 5:
3122: for (i=0; i<mbs; i++) {
3123: ij[0] = 5*i; ij[1] = 5*i + 1; ij[2] = 5*i + 2; ij[3] = 5*i + 3; ij[4] = 5*i + 4;
3124: MatGetValues(A,5,ij,5,ij,diag);
3125: PetscKernel_A_gets_inverse_A_5(diag,ipvt,work,shift);
3126: PetscKernel_A_gets_transpose_A_5(diag);
3127: diag += 25;
3128: }
3129: break;
3130: case 6:
3131: for (i=0; i<mbs; i++) {
3132: 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;
3133: MatGetValues(A,6,ij,6,ij,diag);
3134: PetscKernel_A_gets_inverse_A_6(diag,shift);
3135: PetscKernel_A_gets_transpose_A_6(diag);
3136: diag += 36;
3137: }
3138: break;
3139: case 7:
3140: for (i=0; i<mbs; i++) {
3141: 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;
3142: MatGetValues(A,7,ij,7,ij,diag);
3143: PetscKernel_A_gets_inverse_A_7(diag,shift);
3144: PetscKernel_A_gets_transpose_A_7(diag);
3145: diag += 49;
3146: }
3147: break;
3148: default:
3149: PetscMalloc3(bs,&v_work,bs,&v_pivots,bs,&IJ);
3150: for (i=0; i<mbs; i++) {
3151: for (j=0; j<bs; j++) {
3152: IJ[j] = bs*i + j;
3153: }
3154: MatGetValues(A,bs,IJ,bs,IJ,diag);
3155: PetscKernel_A_gets_inverse_A(bs,diag,v_pivots,v_work);
3156: PetscKernel_A_gets_transpose_A_N(diag,bs);
3157: diag += bs2;
3158: }
3159: PetscFree3(v_work,v_pivots,IJ);
3160: }
3161: a->ibdiagvalid = PETSC_TRUE;
3162: return(0);
3163: }
3167: static PetscErrorCode MatSetRandom_SeqAIJ(Mat x,PetscRandom rctx)
3168: {
3170: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)x->data;
3171: PetscScalar a;
3172: PetscInt m,n,i,j,col;
3175: if (!x->assembled) {
3176: MatGetSize(x,&m,&n);
3177: for (i=0; i<m; i++) {
3178: for (j=0; j<aij->imax[i]; j++) {
3179: PetscRandomGetValue(rctx,&a);
3180: col = (PetscInt)(n*PetscRealPart(a));
3181: MatSetValues(x,1,&i,1,&col,&a,ADD_VALUES);
3182: }
3183: }
3184: } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Not yet coded");
3185: MatAssemblyBegin(x,MAT_FINAL_ASSEMBLY);
3186: MatAssemblyEnd(x,MAT_FINAL_ASSEMBLY);
3187: return(0);
3188: }
3190: /* -------------------------------------------------------------------*/
3191: static struct _MatOps MatOps_Values = { MatSetValues_SeqAIJ,
3192: MatGetRow_SeqAIJ,
3193: MatRestoreRow_SeqAIJ,
3194: MatMult_SeqAIJ,
3195: /* 4*/ MatMultAdd_SeqAIJ,
3196: MatMultTranspose_SeqAIJ,
3197: MatMultTransposeAdd_SeqAIJ,
3198: 0,
3199: 0,
3200: 0,
3201: /* 10*/ 0,
3202: MatLUFactor_SeqAIJ,
3203: 0,
3204: MatSOR_SeqAIJ,
3205: MatTranspose_SeqAIJ,
3206: /*1 5*/ MatGetInfo_SeqAIJ,
3207: MatEqual_SeqAIJ,
3208: MatGetDiagonal_SeqAIJ,
3209: MatDiagonalScale_SeqAIJ,
3210: MatNorm_SeqAIJ,
3211: /* 20*/ 0,
3212: MatAssemblyEnd_SeqAIJ,
3213: MatSetOption_SeqAIJ,
3214: MatZeroEntries_SeqAIJ,
3215: /* 24*/ MatZeroRows_SeqAIJ,
3216: 0,
3217: 0,
3218: 0,
3219: 0,
3220: /* 29*/ MatSetUp_SeqAIJ,
3221: 0,
3222: 0,
3223: 0,
3224: 0,
3225: /* 34*/ MatDuplicate_SeqAIJ,
3226: 0,
3227: 0,
3228: MatILUFactor_SeqAIJ,
3229: 0,
3230: /* 39*/ MatAXPY_SeqAIJ,
3231: MatGetSubMatrices_SeqAIJ,
3232: MatIncreaseOverlap_SeqAIJ,
3233: MatGetValues_SeqAIJ,
3234: MatCopy_SeqAIJ,
3235: /* 44*/ MatGetRowMax_SeqAIJ,
3236: MatScale_SeqAIJ,
3237: 0,
3238: MatDiagonalSet_SeqAIJ,
3239: MatZeroRowsColumns_SeqAIJ,
3240: /* 49*/ MatSetRandom_SeqAIJ,
3241: MatGetRowIJ_SeqAIJ,
3242: MatRestoreRowIJ_SeqAIJ,
3243: MatGetColumnIJ_SeqAIJ,
3244: MatRestoreColumnIJ_SeqAIJ,
3245: /* 54*/ MatFDColoringCreate_SeqXAIJ,
3246: 0,
3247: 0,
3248: MatPermute_SeqAIJ,
3249: 0,
3250: /* 59*/ 0,
3251: MatDestroy_SeqAIJ,
3252: MatView_SeqAIJ,
3253: 0,
3254: MatMatMatMult_SeqAIJ_SeqAIJ_SeqAIJ,
3255: /* 64*/ MatMatMatMultSymbolic_SeqAIJ_SeqAIJ_SeqAIJ,
3256: MatMatMatMultNumeric_SeqAIJ_SeqAIJ_SeqAIJ,
3257: 0,
3258: 0,
3259: 0,
3260: /* 69*/ MatGetRowMaxAbs_SeqAIJ,
3261: MatGetRowMinAbs_SeqAIJ,
3262: 0,
3263: MatSetColoring_SeqAIJ,
3264: 0,
3265: /* 74*/ MatSetValuesAdifor_SeqAIJ,
3266: MatFDColoringApply_AIJ,
3267: 0,
3268: 0,
3269: 0,
3270: /* 79*/ MatFindZeroDiagonals_SeqAIJ,
3271: 0,
3272: 0,
3273: 0,
3274: MatLoad_SeqAIJ,
3275: /* 84*/ MatIsSymmetric_SeqAIJ,
3276: MatIsHermitian_SeqAIJ,
3277: 0,
3278: 0,
3279: 0,
3280: /* 89*/ MatMatMult_SeqAIJ_SeqAIJ,
3281: MatMatMultSymbolic_SeqAIJ_SeqAIJ,
3282: MatMatMultNumeric_SeqAIJ_SeqAIJ,
3283: MatPtAP_SeqAIJ_SeqAIJ,
3284: MatPtAPSymbolic_SeqAIJ_SeqAIJ_DenseAxpy,
3285: /* 94*/ MatPtAPNumeric_SeqAIJ_SeqAIJ,
3286: MatMatTransposeMult_SeqAIJ_SeqAIJ,
3287: MatMatTransposeMultSymbolic_SeqAIJ_SeqAIJ,
3288: MatMatTransposeMultNumeric_SeqAIJ_SeqAIJ,
3289: 0,
3290: /* 99*/ 0,
3291: 0,
3292: 0,
3293: MatConjugate_SeqAIJ,
3294: 0,
3295: /*104*/ MatSetValuesRow_SeqAIJ,
3296: MatRealPart_SeqAIJ,
3297: MatImaginaryPart_SeqAIJ,
3298: 0,
3299: 0,
3300: /*109*/ MatMatSolve_SeqAIJ,
3301: 0,
3302: MatGetRowMin_SeqAIJ,
3303: 0,
3304: MatMissingDiagonal_SeqAIJ,
3305: /*114*/ 0,
3306: 0,
3307: 0,
3308: 0,
3309: 0,
3310: /*119*/ 0,
3311: 0,
3312: 0,
3313: 0,
3314: MatGetMultiProcBlock_SeqAIJ,
3315: /*124*/ MatFindNonzeroRows_SeqAIJ,
3316: MatGetColumnNorms_SeqAIJ,
3317: MatInvertBlockDiagonal_SeqAIJ,
3318: 0,
3319: 0,
3320: /*129*/ 0,
3321: MatTransposeMatMult_SeqAIJ_SeqAIJ,
3322: MatTransposeMatMultSymbolic_SeqAIJ_SeqAIJ,
3323: MatTransposeMatMultNumeric_SeqAIJ_SeqAIJ,
3324: MatTransposeColoringCreate_SeqAIJ,
3325: /*134*/ MatTransColoringApplySpToDen_SeqAIJ,
3326: MatTransColoringApplyDenToSp_SeqAIJ,
3327: MatRARt_SeqAIJ_SeqAIJ,
3328: MatRARtSymbolic_SeqAIJ_SeqAIJ,
3329: MatRARtNumeric_SeqAIJ_SeqAIJ,
3330: /*139*/0,
3331: 0,
3332: 0,
3333: MatFDColoringSetUp_SeqXAIJ,
3334: MatFindOffBlockDiagonalEntries_SeqAIJ
3335: };
3339: PetscErrorCode MatSeqAIJSetColumnIndices_SeqAIJ(Mat mat,PetscInt *indices)
3340: {
3341: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data;
3342: PetscInt i,nz,n;
3345: nz = aij->maxnz;
3346: n = mat->rmap->n;
3347: for (i=0; i<nz; i++) {
3348: aij->j[i] = indices[i];
3349: }
3350: aij->nz = nz;
3351: for (i=0; i<n; i++) {
3352: aij->ilen[i] = aij->imax[i];
3353: }
3354: return(0);
3355: }
3359: /*@
3360: MatSeqAIJSetColumnIndices - Set the column indices for all the rows
3361: in the matrix.
3363: Input Parameters:
3364: + mat - the SeqAIJ matrix
3365: - indices - the column indices
3367: Level: advanced
3369: Notes:
3370: This can be called if you have precomputed the nonzero structure of the
3371: matrix and want to provide it to the matrix object to improve the performance
3372: of the MatSetValues() operation.
3374: You MUST have set the correct numbers of nonzeros per row in the call to
3375: MatCreateSeqAIJ(), and the columns indices MUST be sorted.
3377: MUST be called before any calls to MatSetValues();
3379: The indices should start with zero, not one.
3381: @*/
3382: PetscErrorCode MatSeqAIJSetColumnIndices(Mat mat,PetscInt *indices)
3383: {
3389: PetscUseMethod(mat,"MatSeqAIJSetColumnIndices_C",(Mat,PetscInt*),(mat,indices));
3390: return(0);
3391: }
3393: /* ----------------------------------------------------------------------------------------*/
3397: PetscErrorCode MatStoreValues_SeqAIJ(Mat mat)
3398: {
3399: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data;
3401: size_t nz = aij->i[mat->rmap->n];
3404: if (!aij->nonew) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first");
3406: /* allocate space for values if not already there */
3407: if (!aij->saved_values) {
3408: PetscMalloc1((nz+1),&aij->saved_values);
3409: PetscLogObjectMemory((PetscObject)mat,(nz+1)*sizeof(PetscScalar));
3410: }
3412: /* copy values over */
3413: PetscMemcpy(aij->saved_values,aij->a,nz*sizeof(PetscScalar));
3414: return(0);
3415: }
3419: /*@
3420: MatStoreValues - Stashes a copy of the matrix values; this allows, for
3421: example, reuse of the linear part of a Jacobian, while recomputing the
3422: nonlinear portion.
3424: Collect on Mat
3426: Input Parameters:
3427: . mat - the matrix (currently only AIJ matrices support this option)
3429: Level: advanced
3431: Common Usage, with SNESSolve():
3432: $ Create Jacobian matrix
3433: $ Set linear terms into matrix
3434: $ Apply boundary conditions to matrix, at this time matrix must have
3435: $ final nonzero structure (i.e. setting the nonlinear terms and applying
3436: $ boundary conditions again will not change the nonzero structure
3437: $ MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);
3438: $ MatStoreValues(mat);
3439: $ Call SNESSetJacobian() with matrix
3440: $ In your Jacobian routine
3441: $ MatRetrieveValues(mat);
3442: $ Set nonlinear terms in matrix
3444: Common Usage without SNESSolve(), i.e. when you handle nonlinear solve yourself:
3445: $ // build linear portion of Jacobian
3446: $ MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);
3447: $ MatStoreValues(mat);
3448: $ loop over nonlinear iterations
3449: $ MatRetrieveValues(mat);
3450: $ // call MatSetValues(mat,...) to set nonliner portion of Jacobian
3451: $ // call MatAssemblyBegin/End() on matrix
3452: $ Solve linear system with Jacobian
3453: $ endloop
3455: Notes:
3456: Matrix must already be assemblied before calling this routine
3457: Must set the matrix option MatSetOption(mat,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE); before
3458: calling this routine.
3460: When this is called multiple times it overwrites the previous set of stored values
3461: and does not allocated additional space.
3463: .seealso: MatRetrieveValues()
3465: @*/
3466: PetscErrorCode MatStoreValues(Mat mat)
3467: {
3472: if (!mat->assembled) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix");
3473: if (mat->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
3474: PetscUseMethod(mat,"MatStoreValues_C",(Mat),(mat));
3475: return(0);
3476: }
3480: PetscErrorCode MatRetrieveValues_SeqAIJ(Mat mat)
3481: {
3482: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)mat->data;
3484: PetscInt nz = aij->i[mat->rmap->n];
3487: if (!aij->nonew) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first");
3488: if (!aij->saved_values) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatStoreValues(A);first");
3489: /* copy values over */
3490: PetscMemcpy(aij->a,aij->saved_values,nz*sizeof(PetscScalar));
3491: return(0);
3492: }
3496: /*@
3497: MatRetrieveValues - Retrieves the copy of the matrix values; this allows, for
3498: example, reuse of the linear part of a Jacobian, while recomputing the
3499: nonlinear portion.
3501: Collect on Mat
3503: Input Parameters:
3504: . mat - the matrix (currently on AIJ matrices support this option)
3506: Level: advanced
3508: .seealso: MatStoreValues()
3510: @*/
3511: PetscErrorCode MatRetrieveValues(Mat mat)
3512: {
3517: if (!mat->assembled) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix");
3518: if (mat->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
3519: PetscUseMethod(mat,"MatRetrieveValues_C",(Mat),(mat));
3520: return(0);
3521: }
3524: /* --------------------------------------------------------------------------------*/
3527: /*@C
3528: MatCreateSeqAIJ - Creates a sparse matrix in AIJ (compressed row) format
3529: (the default parallel PETSc format). For good matrix assembly performance
3530: the user should preallocate the matrix storage by setting the parameter nz
3531: (or the array nnz). By setting these parameters accurately, performance
3532: during matrix assembly can be increased by more than a factor of 50.
3534: Collective on MPI_Comm
3536: Input Parameters:
3537: + comm - MPI communicator, set to PETSC_COMM_SELF
3538: . m - number of rows
3539: . n - number of columns
3540: . nz - number of nonzeros per row (same for all rows)
3541: - nnz - array containing the number of nonzeros in the various rows
3542: (possibly different for each row) or NULL
3544: Output Parameter:
3545: . A - the matrix
3547: It is recommended that one use the MatCreate(), MatSetType() and/or MatSetFromOptions(),
3548: MatXXXXSetPreallocation() paradgm instead of this routine directly.
3549: [MatXXXXSetPreallocation() is, for example, MatSeqAIJSetPreallocation]
3551: Notes:
3552: If nnz is given then nz is ignored
3554: The AIJ format (also called the Yale sparse matrix format or
3555: compressed row storage), is fully compatible with standard Fortran 77
3556: storage. That is, the stored row and column indices can begin at
3557: either one (as in Fortran) or zero. See the users' manual for details.
3559: Specify the preallocated storage with either nz or nnz (not both).
3560: Set nz=PETSC_DEFAULT and nnz=NULL for PETSc to control dynamic memory
3561: allocation. For large problems you MUST preallocate memory or you
3562: will get TERRIBLE performance, see the users' manual chapter on matrices.
3564: By default, this format uses inodes (identical nodes) when possible, to
3565: improve numerical efficiency of matrix-vector products and solves. We
3566: search for consecutive rows with the same nonzero structure, thereby
3567: reusing matrix information to achieve increased efficiency.
3569: Options Database Keys:
3570: + -mat_no_inode - Do not use inodes
3571: - -mat_inode_limit <limit> - Sets inode limit (max limit=5)
3573: Level: intermediate
3575: .seealso: MatCreate(), MatCreateAIJ(), MatSetValues(), MatSeqAIJSetColumnIndices(), MatCreateSeqAIJWithArrays()
3577: @*/
3578: PetscErrorCode MatCreateSeqAIJ(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt nz,const PetscInt nnz[],Mat *A)
3579: {
3583: MatCreate(comm,A);
3584: MatSetSizes(*A,m,n,m,n);
3585: MatSetType(*A,MATSEQAIJ);
3586: MatSeqAIJSetPreallocation_SeqAIJ(*A,nz,nnz);
3587: return(0);
3588: }
3592: /*@C
3593: MatSeqAIJSetPreallocation - For good matrix assembly performance
3594: the user should preallocate the matrix storage by setting the parameter nz
3595: (or the array nnz). By setting these parameters accurately, performance
3596: during matrix assembly can be increased by more than a factor of 50.
3598: Collective on MPI_Comm
3600: Input Parameters:
3601: + B - The matrix
3602: . nz - number of nonzeros per row (same for all rows)
3603: - nnz - array containing the number of nonzeros in the various rows
3604: (possibly different for each row) or NULL
3606: Notes:
3607: If nnz is given then nz is ignored
3609: The AIJ format (also called the Yale sparse matrix format or
3610: compressed row storage), is fully compatible with standard Fortran 77
3611: storage. That is, the stored row and column indices can begin at
3612: either one (as in Fortran) or zero. See the users' manual for details.
3614: Specify the preallocated storage with either nz or nnz (not both).
3615: Set nz=PETSC_DEFAULT and nnz=NULL for PETSc to control dynamic memory
3616: allocation. For large problems you MUST preallocate memory or you
3617: will get TERRIBLE performance, see the users' manual chapter on matrices.
3619: You can call MatGetInfo() to get information on how effective the preallocation was;
3620: for example the fields mallocs,nz_allocated,nz_used,nz_unneeded;
3621: You can also run with the option -info and look for messages with the string
3622: malloc in them to see if additional memory allocation was needed.
3624: Developers: Use nz of MAT_SKIP_ALLOCATION to not allocate any space for the matrix
3625: entries or columns indices
3627: By default, this format uses inodes (identical nodes) when possible, to
3628: improve numerical efficiency of matrix-vector products and solves. We
3629: search for consecutive rows with the same nonzero structure, thereby
3630: reusing matrix information to achieve increased efficiency.
3632: Options Database Keys:
3633: + -mat_no_inode - Do not use inodes
3634: . -mat_inode_limit <limit> - Sets inode limit (max limit=5)
3635: - -mat_aij_oneindex - Internally use indexing starting at 1
3636: rather than 0. Note that when calling MatSetValues(),
3637: the user still MUST index entries starting at 0!
3639: Level: intermediate
3641: .seealso: MatCreate(), MatCreateAIJ(), MatSetValues(), MatSeqAIJSetColumnIndices(), MatCreateSeqAIJWithArrays(), MatGetInfo()
3643: @*/
3644: PetscErrorCode MatSeqAIJSetPreallocation(Mat B,PetscInt nz,const PetscInt nnz[])
3645: {
3651: PetscTryMethod(B,"MatSeqAIJSetPreallocation_C",(Mat,PetscInt,const PetscInt[]),(B,nz,nnz));
3652: return(0);
3653: }
3657: PetscErrorCode MatSeqAIJSetPreallocation_SeqAIJ(Mat B,PetscInt nz,const PetscInt *nnz)
3658: {
3659: Mat_SeqAIJ *b;
3660: PetscBool skipallocation = PETSC_FALSE,realalloc = PETSC_FALSE;
3662: PetscInt i;
3665: if (nz >= 0 || nnz) realalloc = PETSC_TRUE;
3666: if (nz == MAT_SKIP_ALLOCATION) {
3667: skipallocation = PETSC_TRUE;
3668: nz = 0;
3669: }
3671: PetscLayoutSetUp(B->rmap);
3672: PetscLayoutSetUp(B->cmap);
3674: if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 5;
3675: if (nz < 0) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"nz cannot be less than 0: value %D",nz);
3676: if (nnz) {
3677: for (i=0; i<B->rmap->n; i++) {
3678: 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]);
3679: 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);
3680: }
3681: }
3683: B->preallocated = PETSC_TRUE;
3685: b = (Mat_SeqAIJ*)B->data;
3687: if (!skipallocation) {
3688: if (!b->imax) {
3689: PetscMalloc2(B->rmap->n,&b->imax,B->rmap->n,&b->ilen);
3690: PetscLogObjectMemory((PetscObject)B,2*B->rmap->n*sizeof(PetscInt));
3691: }
3692: if (!nnz) {
3693: if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 10;
3694: else if (nz < 0) nz = 1;
3695: for (i=0; i<B->rmap->n; i++) b->imax[i] = nz;
3696: nz = nz*B->rmap->n;
3697: } else {
3698: nz = 0;
3699: for (i=0; i<B->rmap->n; i++) {b->imax[i] = nnz[i]; nz += nnz[i];}
3700: }
3701: /* b->ilen will count nonzeros in each row so far. */
3702: for (i=0; i<B->rmap->n; i++) b->ilen[i] = 0;
3704: /* allocate the matrix space */
3705: MatSeqXAIJFreeAIJ(B,&b->a,&b->j,&b->i);
3706: PetscMalloc3(nz,&b->a,nz,&b->j,B->rmap->n+1,&b->i);
3707: PetscLogObjectMemory((PetscObject)B,(B->rmap->n+1)*sizeof(PetscInt)+nz*(sizeof(PetscScalar)+sizeof(PetscInt)));
3708: b->i[0] = 0;
3709: for (i=1; i<B->rmap->n+1; i++) {
3710: b->i[i] = b->i[i-1] + b->imax[i-1];
3711: }
3712: b->singlemalloc = PETSC_TRUE;
3713: b->free_a = PETSC_TRUE;
3714: b->free_ij = PETSC_TRUE;
3715: #if defined(PETSC_THREADCOMM_ACTIVE)
3716: MatZeroEntries_SeqAIJ(B);
3717: #endif
3718: } else {
3719: b->free_a = PETSC_FALSE;
3720: b->free_ij = PETSC_FALSE;
3721: }
3723: b->nz = 0;
3724: b->maxnz = nz;
3725: B->info.nz_unneeded = (double)b->maxnz;
3726: if (realalloc) {
3727: MatSetOption(B,MAT_NEW_NONZERO_ALLOCATION_ERR,PETSC_TRUE);
3728: }
3729: return(0);
3730: }
3732: #undef __FUNCT__
3734: /*@
3735: MatSeqAIJSetPreallocationCSR - Allocates memory for a sparse sequential matrix in AIJ format.
3737: Input Parameters:
3738: + B - the matrix
3739: . i - the indices into j for the start of each row (starts with zero)
3740: . j - the column indices for each row (starts with zero) these must be sorted for each row
3741: - v - optional values in the matrix
3743: Level: developer
3745: The i,j,v values are COPIED with this routine; to avoid the copy use MatCreateSeqAIJWithArrays()
3747: .keywords: matrix, aij, compressed row, sparse, sequential
3749: .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatSeqAIJSetPreallocation(), MatCreateSeqAIJ(), SeqAIJ
3750: @*/
3751: PetscErrorCode MatSeqAIJSetPreallocationCSR(Mat B,const PetscInt i[],const PetscInt j[],const PetscScalar v[])
3752: {
3758: PetscTryMethod(B,"MatSeqAIJSetPreallocationCSR_C",(Mat,const PetscInt[],const PetscInt[],const PetscScalar[]),(B,i,j,v));
3759: return(0);
3760: }
3762: #undef __FUNCT__
3764: PetscErrorCode MatSeqAIJSetPreallocationCSR_SeqAIJ(Mat B,const PetscInt Ii[],const PetscInt J[],const PetscScalar v[])
3765: {
3766: PetscInt i;
3767: PetscInt m,n;
3768: PetscInt nz;
3769: PetscInt *nnz, nz_max = 0;
3770: PetscScalar *values;
3774: if (Ii[0]) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE, "Ii[0] must be 0 it is %D", Ii[0]);
3776: PetscLayoutSetUp(B->rmap);
3777: PetscLayoutSetUp(B->cmap);
3779: MatGetSize(B, &m, &n);
3780: PetscMalloc1((m+1), &nnz);
3781: for (i = 0; i < m; i++) {
3782: nz = Ii[i+1]- Ii[i];
3783: nz_max = PetscMax(nz_max, nz);
3784: if (nz < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE, "Local row %D has a negative number of columns %D", i, nnz);
3785: nnz[i] = nz;
3786: }
3787: MatSeqAIJSetPreallocation(B, 0, nnz);
3788: PetscFree(nnz);
3790: if (v) {
3791: values = (PetscScalar*) v;
3792: } else {
3793: PetscCalloc1(nz_max, &values);
3794: }
3796: for (i = 0; i < m; i++) {
3797: nz = Ii[i+1] - Ii[i];
3798: MatSetValues_SeqAIJ(B, 1, &i, nz, J+Ii[i], values + (v ? Ii[i] : 0), INSERT_VALUES);
3799: }
3801: MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);
3802: MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);
3804: if (!v) {
3805: PetscFree(values);
3806: }
3807: MatSetOption(B,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_TRUE);
3808: return(0);
3809: }
3811: #include <../src/mat/impls/dense/seq/dense.h>
3812: #include <petsc-private/kernels/petscaxpy.h>
3816: /*
3817: Computes (B'*A')' since computing B*A directly is untenable
3819: n p p
3820: ( ) ( ) ( )
3821: m ( A ) * n ( B ) = m ( C )
3822: ( ) ( ) ( )
3824: */
3825: PetscErrorCode MatMatMultNumeric_SeqDense_SeqAIJ(Mat A,Mat B,Mat C)
3826: {
3827: PetscErrorCode ierr;
3828: Mat_SeqDense *sub_a = (Mat_SeqDense*)A->data;
3829: Mat_SeqAIJ *sub_b = (Mat_SeqAIJ*)B->data;
3830: Mat_SeqDense *sub_c = (Mat_SeqDense*)C->data;
3831: PetscInt i,n,m,q,p;
3832: const PetscInt *ii,*idx;
3833: const PetscScalar *b,*a,*a_q;
3834: PetscScalar *c,*c_q;
3837: m = A->rmap->n;
3838: n = A->cmap->n;
3839: p = B->cmap->n;
3840: a = sub_a->v;
3841: b = sub_b->a;
3842: c = sub_c->v;
3843: PetscMemzero(c,m*p*sizeof(PetscScalar));
3845: ii = sub_b->i;
3846: idx = sub_b->j;
3847: for (i=0; i<n; i++) {
3848: q = ii[i+1] - ii[i];
3849: while (q-->0) {
3850: c_q = c + m*(*idx);
3851: a_q = a + m*i;
3852: PetscKernelAXPY(c_q,*b,a_q,m);
3853: idx++;
3854: b++;
3855: }
3856: }
3857: return(0);
3858: }
3862: PetscErrorCode MatMatMultSymbolic_SeqDense_SeqAIJ(Mat A,Mat B,PetscReal fill,Mat *C)
3863: {
3865: PetscInt m=A->rmap->n,n=B->cmap->n;
3866: Mat Cmat;
3869: 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);
3870: MatCreate(PetscObjectComm((PetscObject)A),&Cmat);
3871: MatSetSizes(Cmat,m,n,m,n);
3872: MatSetBlockSizesFromMats(Cmat,A,B);
3873: MatSetType(Cmat,MATSEQDENSE);
3874: MatSeqDenseSetPreallocation(Cmat,NULL);
3876: Cmat->ops->matmultnumeric = MatMatMultNumeric_SeqDense_SeqAIJ;
3878: *C = Cmat;
3879: return(0);
3880: }
3882: /* ----------------------------------------------------------------*/
3885: PetscErrorCode MatMatMult_SeqDense_SeqAIJ(Mat A,Mat B,MatReuse scall,PetscReal fill,Mat *C)
3886: {
3890: if (scall == MAT_INITIAL_MATRIX) {
3891: PetscLogEventBegin(MAT_MatMultSymbolic,A,B,0,0);
3892: MatMatMultSymbolic_SeqDense_SeqAIJ(A,B,fill,C);
3893: PetscLogEventEnd(MAT_MatMultSymbolic,A,B,0,0);
3894: }
3895: PetscLogEventBegin(MAT_MatMultNumeric,A,B,0,0);
3896: MatMatMultNumeric_SeqDense_SeqAIJ(A,B,*C);
3897: PetscLogEventEnd(MAT_MatMultNumeric,A,B,0,0);
3898: return(0);
3899: }
3902: /*MC
3903: MATSEQAIJ - MATSEQAIJ = "seqaij" - A matrix type to be used for sequential sparse matrices,
3904: based on compressed sparse row format.
3906: Options Database Keys:
3907: . -mat_type seqaij - sets the matrix type to "seqaij" during a call to MatSetFromOptions()
3909: Level: beginner
3911: .seealso: MatCreateSeqAIJ(), MatSetFromOptions(), MatSetType(), MatCreate(), MatType
3912: M*/
3914: /*MC
3915: MATAIJ - MATAIJ = "aij" - A matrix type to be used for sparse matrices.
3917: This matrix type is identical to MATSEQAIJ when constructed with a single process communicator,
3918: and MATMPIAIJ otherwise. As a result, for single process communicators,
3919: MatSeqAIJSetPreallocation is supported, and similarly MatMPIAIJSetPreallocation is supported
3920: for communicators controlling multiple processes. It is recommended that you call both of
3921: the above preallocation routines for simplicity.
3923: Options Database Keys:
3924: . -mat_type aij - sets the matrix type to "aij" during a call to MatSetFromOptions()
3926: Developer Notes: Subclasses include MATAIJCUSP, MATAIJPERM, MATAIJCRL, and also automatically switches over to use inodes when
3927: enough exist.
3929: Level: beginner
3931: .seealso: MatCreateAIJ(), MatCreateSeqAIJ(), MATSEQAIJ,MATMPIAIJ
3932: M*/
3934: /*MC
3935: MATAIJCRL - MATAIJCRL = "aijcrl" - A matrix type to be used for sparse matrices.
3937: This matrix type is identical to MATSEQAIJCRL when constructed with a single process communicator,
3938: and MATMPIAIJCRL otherwise. As a result, for single process communicators,
3939: MatSeqAIJSetPreallocation() is supported, and similarly MatMPIAIJSetPreallocation() is supported
3940: for communicators controlling multiple processes. It is recommended that you call both of
3941: the above preallocation routines for simplicity.
3943: Options Database Keys:
3944: . -mat_type aijcrl - sets the matrix type to "aijcrl" during a call to MatSetFromOptions()
3946: Level: beginner
3948: .seealso: MatCreateMPIAIJCRL,MATSEQAIJCRL,MATMPIAIJCRL, MATSEQAIJCRL, MATMPIAIJCRL
3949: M*/
3951: #if defined(PETSC_HAVE_PASTIX)
3952: PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_pastix(Mat,MatFactorType,Mat*);
3953: #endif
3954: #if defined(PETSC_HAVE_ESSL) && !defined(PETSC_USE_COMPLEX) && !defined(PETSC_USE_REAL_SINGLE) && !defined(PETSC_USE_REAL___FLOAT128)
3955: PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_essl(Mat,MatFactorType,Mat*);
3956: #endif
3957: PETSC_EXTERN PetscErrorCode MatConvert_SeqAIJ_SeqAIJCRL(Mat,MatType,MatReuse,Mat*);
3958: PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_petsc(Mat,MatFactorType,Mat*);
3959: PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_bas(Mat,MatFactorType,Mat*);
3960: extern PetscErrorCode MatGetFactorAvailable_seqaij_petsc(Mat,MatFactorType,PetscBool*);
3961: #if defined(PETSC_HAVE_MUMPS)
3962: PETSC_EXTERN PetscErrorCode MatGetFactor_aij_mumps(Mat,MatFactorType,Mat*);
3963: #endif
3964: #if defined(PETSC_HAVE_SUPERLU)
3965: PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_superlu(Mat,MatFactorType,Mat*);
3966: #endif
3967: #if defined(PETSC_HAVE_MKL_PARDISO)
3968: PETSC_EXTERN PetscErrorCode MatGetFactor_aij_mkl_pardiso(Mat,MatFactorType,Mat*);
3969: #endif
3970: #if defined(PETSC_HAVE_SUPERLU_DIST)
3971: PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_superlu_dist(Mat,MatFactorType,Mat*);
3972: #endif
3973: #if defined(PETSC_HAVE_SUITESPARSE)
3974: PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_umfpack(Mat,MatFactorType,Mat*);
3975: #endif
3976: #if defined(PETSC_HAVE_SUITESPARSE)
3977: PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_cholmod(Mat,MatFactorType,Mat*);
3978: #endif
3979: #if defined(PETSC_HAVE_SUITESPARSE)
3980: PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_klu(Mat,MatFactorType,Mat*);
3981: #endif
3982: #if defined(PETSC_HAVE_LUSOL)
3983: PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_lusol(Mat,MatFactorType,Mat*);
3984: #endif
3985: #if defined(PETSC_HAVE_MATLAB_ENGINE)
3986: PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_matlab(Mat,MatFactorType,Mat*);
3987: extern PetscErrorCode MatlabEnginePut_SeqAIJ(PetscObject,void*);
3988: extern PetscErrorCode MatlabEngineGet_SeqAIJ(PetscObject,void*);
3989: #endif
3990: #if defined(PETSC_HAVE_CLIQUE)
3991: PETSC_EXTERN PetscErrorCode MatGetFactor_aij_clique(Mat,MatFactorType,Mat*);
3992: #endif
3997: /*@C
3998: MatSeqAIJGetArray - gives access to the array where the data for a SeqSeqAIJ matrix is stored
4000: Not Collective
4002: Input Parameter:
4003: . mat - a MATSEQDENSE matrix
4005: Output Parameter:
4006: . array - pointer to the data
4008: Level: intermediate
4010: .seealso: MatSeqAIJRestoreArray(), MatSeqAIJGetArrayF90()
4011: @*/
4012: PetscErrorCode MatSeqAIJGetArray(Mat A,PetscScalar **array)
4013: {
4017: PetscUseMethod(A,"MatSeqAIJGetArray_C",(Mat,PetscScalar**),(A,array));
4018: return(0);
4019: }
4023: /*@C
4024: MatSeqAIJGetMaxRowNonzeros - returns the maximum number of nonzeros in any row
4026: Not Collective
4028: Input Parameter:
4029: . mat - a MATSEQDENSE matrix
4031: Output Parameter:
4032: . nz - the maximum number of nonzeros in any row
4034: Level: intermediate
4036: .seealso: MatSeqAIJRestoreArray(), MatSeqAIJGetArrayF90()
4037: @*/
4038: PetscErrorCode MatSeqAIJGetMaxRowNonzeros(Mat A,PetscInt *nz)
4039: {
4040: Mat_SeqAIJ *aij = (Mat_SeqAIJ*)A->data;
4043: *nz = aij->rmax;
4044: return(0);
4045: }
4049: /*@C
4050: MatSeqAIJRestoreArray - returns access to the array where the data for a SeqSeqAIJ matrix is stored obtained by MatSeqAIJGetArray()
4052: Not Collective
4054: Input Parameters:
4055: . mat - a MATSEQDENSE matrix
4056: . array - pointer to the data
4058: Level: intermediate
4060: .seealso: MatSeqAIJGetArray(), MatSeqAIJRestoreArrayF90()
4061: @*/
4062: PetscErrorCode MatSeqAIJRestoreArray(Mat A,PetscScalar **array)
4063: {
4067: PetscUseMethod(A,"MatSeqAIJRestoreArray_C",(Mat,PetscScalar**),(A,array));
4068: return(0);
4069: }
4073: PETSC_EXTERN PetscErrorCode MatCreate_SeqAIJ(Mat B)
4074: {
4075: Mat_SeqAIJ *b;
4077: PetscMPIInt size;
4080: MPI_Comm_size(PetscObjectComm((PetscObject)B),&size);
4081: if (size > 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Comm must be of size 1");
4083: PetscNewLog(B,&b);
4085: B->data = (void*)b;
4087: PetscMemcpy(B->ops,&MatOps_Values,sizeof(struct _MatOps));
4089: b->row = 0;
4090: b->col = 0;
4091: b->icol = 0;
4092: b->reallocs = 0;
4093: b->ignorezeroentries = PETSC_FALSE;
4094: b->roworiented = PETSC_TRUE;
4095: b->nonew = 0;
4096: b->diag = 0;
4097: b->solve_work = 0;
4098: B->spptr = 0;
4099: b->saved_values = 0;
4100: b->idiag = 0;
4101: b->mdiag = 0;
4102: b->ssor_work = 0;
4103: b->omega = 1.0;
4104: b->fshift = 0.0;
4105: b->idiagvalid = PETSC_FALSE;
4106: b->ibdiagvalid = PETSC_FALSE;
4107: b->keepnonzeropattern = PETSC_FALSE;
4108: b->xtoy = 0;
4109: b->XtoY = 0;
4111: PetscObjectChangeTypeName((PetscObject)B,MATSEQAIJ);
4112: PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJGetArray_C",MatSeqAIJGetArray_SeqAIJ);
4113: PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJRestoreArray_C",MatSeqAIJRestoreArray_SeqAIJ);
4115: #if defined(PETSC_HAVE_MATLAB_ENGINE)
4116: PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_matlab_C",MatGetFactor_seqaij_matlab);
4117: PetscObjectComposeFunction((PetscObject)B,"PetscMatlabEnginePut_C",MatlabEnginePut_SeqAIJ);
4118: PetscObjectComposeFunction((PetscObject)B,"PetscMatlabEngineGet_C",MatlabEngineGet_SeqAIJ);
4119: #endif
4120: #if defined(PETSC_HAVE_PASTIX)
4121: PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_pastix_C",MatGetFactor_seqaij_pastix);
4122: #endif
4123: #if defined(PETSC_HAVE_ESSL) && !defined(PETSC_USE_COMPLEX) && !defined(PETSC_USE_REAL_SINGLE) && !defined(PETSC_USE_REAL___FLOAT128)
4124: PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_essl_C",MatGetFactor_seqaij_essl);
4125: #endif
4126: #if defined(PETSC_HAVE_SUPERLU)
4127: PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_superlu_C",MatGetFactor_seqaij_superlu);
4128: #endif
4129: #if defined(PETSC_HAVE_MKL_PARDISO)
4130: PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_mkl_pardiso_C",MatGetFactor_aij_mkl_pardiso);
4131: #endif
4132: #if defined(PETSC_HAVE_SUPERLU_DIST)
4133: PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_superlu_dist_C",MatGetFactor_seqaij_superlu_dist);
4134: #endif
4135: #if defined(PETSC_HAVE_MUMPS)
4136: PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_mumps_C",MatGetFactor_aij_mumps);
4137: #endif
4138: #if defined(PETSC_HAVE_SUITESPARSE)
4139: PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_umfpack_C",MatGetFactor_seqaij_umfpack);
4140: #endif
4141: #if defined(PETSC_HAVE_SUITESPARSE)
4142: PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_cholmod_C",MatGetFactor_seqaij_cholmod);
4143: #endif
4144: #if defined(PETSC_HAVE_SUITESPARSE)
4145: PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_klu_C",MatGetFactor_seqaij_klu);
4146: #endif
4147: #if defined(PETSC_HAVE_LUSOL)
4148: PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_lusol_C",MatGetFactor_seqaij_lusol);
4149: #endif
4150: #if defined(PETSC_HAVE_CLIQUE)
4151: PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_clique_C",MatGetFactor_aij_clique);
4152: #endif
4154: PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_petsc_C",MatGetFactor_seqaij_petsc);
4155: PetscObjectComposeFunction((PetscObject)B,"MatGetFactorAvailable_petsc_C",MatGetFactorAvailable_seqaij_petsc);
4156: PetscObjectComposeFunction((PetscObject)B,"MatGetFactor_bas_C",MatGetFactor_seqaij_bas);
4157: PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetColumnIndices_C",MatSeqAIJSetColumnIndices_SeqAIJ);
4158: PetscObjectComposeFunction((PetscObject)B,"MatStoreValues_C",MatStoreValues_SeqAIJ);
4159: PetscObjectComposeFunction((PetscObject)B,"MatRetrieveValues_C",MatRetrieveValues_SeqAIJ);
4160: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqsbaij_C",MatConvert_SeqAIJ_SeqSBAIJ);
4161: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqbaij_C",MatConvert_SeqAIJ_SeqBAIJ);
4162: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijperm_C",MatConvert_SeqAIJ_SeqAIJPERM);
4163: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqaij_seqaijcrl_C",MatConvert_SeqAIJ_SeqAIJCRL);
4164: PetscObjectComposeFunction((PetscObject)B,"MatIsTranspose_C",MatIsTranspose_SeqAIJ);
4165: PetscObjectComposeFunction((PetscObject)B,"MatIsHermitianTranspose_C",MatIsTranspose_SeqAIJ);
4166: PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetPreallocation_C",MatSeqAIJSetPreallocation_SeqAIJ);
4167: PetscObjectComposeFunction((PetscObject)B,"MatSeqAIJSetPreallocationCSR_C",MatSeqAIJSetPreallocationCSR_SeqAIJ);
4168: PetscObjectComposeFunction((PetscObject)B,"MatReorderForNonzeroDiagonal_C",MatReorderForNonzeroDiagonal_SeqAIJ);
4169: PetscObjectComposeFunction((PetscObject)B,"MatMatMult_seqdense_seqaij_C",MatMatMult_SeqDense_SeqAIJ);
4170: PetscObjectComposeFunction((PetscObject)B,"MatMatMultSymbolic_seqdense_seqaij_C",MatMatMultSymbolic_SeqDense_SeqAIJ);
4171: PetscObjectComposeFunction((PetscObject)B,"MatMatMultNumeric_seqdense_seqaij_C",MatMatMultNumeric_SeqDense_SeqAIJ);
4172: MatCreate_SeqAIJ_Inode(B);
4173: PetscObjectChangeTypeName((PetscObject)B,MATSEQAIJ);
4174: return(0);
4175: }
4179: /*
4180: Given a matrix generated with MatGetFactor() duplicates all the information in A into B
4181: */
4182: PetscErrorCode MatDuplicateNoCreate_SeqAIJ(Mat C,Mat A,MatDuplicateOption cpvalues,PetscBool mallocmatspace)
4183: {
4184: Mat_SeqAIJ *c,*a = (Mat_SeqAIJ*)A->data;
4186: PetscInt i,m = A->rmap->n;
4189: c = (Mat_SeqAIJ*)C->data;
4191: C->factortype = A->factortype;
4192: c->row = 0;
4193: c->col = 0;
4194: c->icol = 0;
4195: c->reallocs = 0;
4197: C->assembled = PETSC_TRUE;
4199: PetscLayoutReference(A->rmap,&C->rmap);
4200: PetscLayoutReference(A->cmap,&C->cmap);
4202: PetscMalloc2(m,&c->imax,m,&c->ilen);
4203: PetscLogObjectMemory((PetscObject)C, 2*m*sizeof(PetscInt));
4204: for (i=0; i<m; i++) {
4205: c->imax[i] = a->imax[i];
4206: c->ilen[i] = a->ilen[i];
4207: }
4209: /* allocate the matrix space */
4210: if (mallocmatspace) {
4211: PetscMalloc3(a->i[m],&c->a,a->i[m],&c->j,m+1,&c->i);
4212: PetscLogObjectMemory((PetscObject)C, a->i[m]*(sizeof(PetscScalar)+sizeof(PetscInt))+(m+1)*sizeof(PetscInt));
4214: c->singlemalloc = PETSC_TRUE;
4216: PetscMemcpy(c->i,a->i,(m+1)*sizeof(PetscInt));
4217: if (m > 0) {
4218: PetscMemcpy(c->j,a->j,(a->i[m])*sizeof(PetscInt));
4219: if (cpvalues == MAT_COPY_VALUES) {
4220: PetscMemcpy(c->a,a->a,(a->i[m])*sizeof(PetscScalar));
4221: } else {
4222: PetscMemzero(c->a,(a->i[m])*sizeof(PetscScalar));
4223: }
4224: }
4225: }
4227: c->ignorezeroentries = a->ignorezeroentries;
4228: c->roworiented = a->roworiented;
4229: c->nonew = a->nonew;
4230: if (a->diag) {
4231: PetscMalloc1((m+1),&c->diag);
4232: PetscLogObjectMemory((PetscObject)C,(m+1)*sizeof(PetscInt));
4233: for (i=0; i<m; i++) {
4234: c->diag[i] = a->diag[i];
4235: }
4236: } else c->diag = 0;
4238: c->solve_work = 0;
4239: c->saved_values = 0;
4240: c->idiag = 0;
4241: c->ssor_work = 0;
4242: c->keepnonzeropattern = a->keepnonzeropattern;
4243: c->free_a = PETSC_TRUE;
4244: c->free_ij = PETSC_TRUE;
4245: c->xtoy = 0;
4246: c->XtoY = 0;
4248: c->rmax = a->rmax;
4249: c->nz = a->nz;
4250: c->maxnz = a->nz; /* Since we allocate exactly the right amount */
4251: C->preallocated = PETSC_TRUE;
4253: c->compressedrow.use = a->compressedrow.use;
4254: c->compressedrow.nrows = a->compressedrow.nrows;
4255: if (a->compressedrow.use) {
4256: i = a->compressedrow.nrows;
4257: PetscMalloc2(i+1,&c->compressedrow.i,i,&c->compressedrow.rindex);
4258: PetscMemcpy(c->compressedrow.i,a->compressedrow.i,(i+1)*sizeof(PetscInt));
4259: PetscMemcpy(c->compressedrow.rindex,a->compressedrow.rindex,i*sizeof(PetscInt));
4260: } else {
4261: c->compressedrow.use = PETSC_FALSE;
4262: c->compressedrow.i = NULL;
4263: c->compressedrow.rindex = NULL;
4264: }
4265: C->nonzerostate = A->nonzerostate;
4267: MatDuplicate_SeqAIJ_Inode(A,cpvalues,&C);
4268: PetscFunctionListDuplicate(((PetscObject)A)->qlist,&((PetscObject)C)->qlist);
4269: return(0);
4270: }
4274: PetscErrorCode MatDuplicate_SeqAIJ(Mat A,MatDuplicateOption cpvalues,Mat *B)
4275: {
4279: MatCreate(PetscObjectComm((PetscObject)A),B);
4280: MatSetSizes(*B,A->rmap->n,A->cmap->n,A->rmap->n,A->cmap->n);
4281: if (!(A->rmap->n % A->rmap->bs) && !(A->cmap->n % A->cmap->bs)) {
4282: MatSetBlockSizesFromMats(*B,A,A);
4283: }
4284: MatSetType(*B,((PetscObject)A)->type_name);
4285: MatDuplicateNoCreate_SeqAIJ(*B,A,cpvalues,PETSC_TRUE);
4286: return(0);
4287: }
4291: PetscErrorCode MatLoad_SeqAIJ(Mat newMat, PetscViewer viewer)
4292: {
4293: Mat_SeqAIJ *a;
4295: PetscInt i,sum,nz,header[4],*rowlengths = 0,M,N,rows,cols;
4296: int fd;
4297: PetscMPIInt size;
4298: MPI_Comm comm;
4299: PetscInt bs = 1;
4302: PetscObjectGetComm((PetscObject)viewer,&comm);
4303: MPI_Comm_size(comm,&size);
4304: if (size > 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"view must have one processor");
4306: PetscOptionsBegin(comm,NULL,"Options for loading SEQAIJ matrix","Mat");
4307: PetscOptionsInt("-matload_block_size","Set the blocksize used to store the matrix","MatLoad",bs,&bs,NULL);
4308: PetscOptionsEnd();
4309: if (bs > 1) {MatSetBlockSize(newMat,bs);}
4311: PetscViewerBinaryGetDescriptor(viewer,&fd);
4312: PetscBinaryRead(fd,header,4,PETSC_INT);
4313: if (header[0] != MAT_FILE_CLASSID) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"not matrix object in file");
4314: M = header[1]; N = header[2]; nz = header[3];
4316: if (nz < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Matrix stored in special format on disk,cannot load as SeqAIJ");
4318: /* read in row lengths */
4319: PetscMalloc1(M,&rowlengths);
4320: PetscBinaryRead(fd,rowlengths,M,PETSC_INT);
4322: /* check if sum of rowlengths is same as nz */
4323: for (i=0,sum=0; i< M; i++) sum +=rowlengths[i];
4324: if (sum != nz) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_FILE_READ,"Inconsistant matrix data in file. no-nonzeros = %dD, sum-row-lengths = %D\n",nz,sum);
4326: /* set global size if not set already*/
4327: if (newMat->rmap->n < 0 && newMat->rmap->N < 0 && newMat->cmap->n < 0 && newMat->cmap->N < 0) {
4328: MatSetSizes(newMat,PETSC_DECIDE,PETSC_DECIDE,M,N);
4329: } else {
4330: /* if sizes and type are already set, check if the vector global sizes are correct */
4331: MatGetSize(newMat,&rows,&cols);
4332: if (rows < 0 && cols < 0) { /* user might provide local size instead of global size */
4333: MatGetLocalSize(newMat,&rows,&cols);
4334: }
4335: 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);
4336: }
4337: MatSeqAIJSetPreallocation_SeqAIJ(newMat,0,rowlengths);
4338: a = (Mat_SeqAIJ*)newMat->data;
4340: PetscBinaryRead(fd,a->j,nz,PETSC_INT);
4342: /* read in nonzero values */
4343: PetscBinaryRead(fd,a->a,nz,PETSC_SCALAR);
4345: /* set matrix "i" values */
4346: a->i[0] = 0;
4347: for (i=1; i<= M; i++) {
4348: a->i[i] = a->i[i-1] + rowlengths[i-1];
4349: a->ilen[i-1] = rowlengths[i-1];
4350: }
4351: PetscFree(rowlengths);
4353: MatAssemblyBegin(newMat,MAT_FINAL_ASSEMBLY);
4354: MatAssemblyEnd(newMat,MAT_FINAL_ASSEMBLY);
4355: return(0);
4356: }
4360: PetscErrorCode MatEqual_SeqAIJ(Mat A,Mat B,PetscBool * flg)
4361: {
4362: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*b = (Mat_SeqAIJ*)B->data;
4364: #if defined(PETSC_USE_COMPLEX)
4365: PetscInt k;
4366: #endif
4369: /* If the matrix dimensions are not equal,or no of nonzeros */
4370: if ((A->rmap->n != B->rmap->n) || (A->cmap->n != B->cmap->n) ||(a->nz != b->nz)) {
4371: *flg = PETSC_FALSE;
4372: return(0);
4373: }
4375: /* if the a->i are the same */
4376: PetscMemcmp(a->i,b->i,(A->rmap->n+1)*sizeof(PetscInt),flg);
4377: if (!*flg) return(0);
4379: /* if a->j are the same */
4380: PetscMemcmp(a->j,b->j,(a->nz)*sizeof(PetscInt),flg);
4381: if (!*flg) return(0);
4383: /* if a->a are the same */
4384: #if defined(PETSC_USE_COMPLEX)
4385: for (k=0; k<a->nz; k++) {
4386: if (PetscRealPart(a->a[k]) != PetscRealPart(b->a[k]) || PetscImaginaryPart(a->a[k]) != PetscImaginaryPart(b->a[k])) {
4387: *flg = PETSC_FALSE;
4388: return(0);
4389: }
4390: }
4391: #else
4392: PetscMemcmp(a->a,b->a,(a->nz)*sizeof(PetscScalar),flg);
4393: #endif
4394: return(0);
4395: }
4399: /*@
4400: MatCreateSeqAIJWithArrays - Creates an sequential AIJ matrix using matrix elements (in CSR format)
4401: provided by the user.
4403: Collective on MPI_Comm
4405: Input Parameters:
4406: + comm - must be an MPI communicator of size 1
4407: . m - number of rows
4408: . n - number of columns
4409: . i - row indices
4410: . j - column indices
4411: - a - matrix values
4413: Output Parameter:
4414: . mat - the matrix
4416: Level: intermediate
4418: Notes:
4419: The i, j, and a arrays are not copied by this routine, the user must free these arrays
4420: once the matrix is destroyed and not before
4422: You cannot set new nonzero locations into this matrix, that will generate an error.
4424: The i and j indices are 0 based
4426: The format which is used for the sparse matrix input, is equivalent to a
4427: row-major ordering.. i.e for the following matrix, the input data expected is
4428: as shown:
4430: 1 0 0
4431: 2 0 3
4432: 4 5 6
4434: i = {0,1,3,6} [size = nrow+1 = 3+1]
4435: j = {0,0,2,0,1,2} [size = nz = 6]; values must be sorted for each row
4436: v = {1,2,3,4,5,6} [size = nz = 6]
4439: .seealso: MatCreate(), MatCreateAIJ(), MatCreateSeqAIJ(), MatCreateMPIAIJWithArrays(), MatMPIAIJSetPreallocationCSR()
4441: @*/
4442: PetscErrorCode MatCreateSeqAIJWithArrays(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt *i,PetscInt *j,PetscScalar *a,Mat *mat)
4443: {
4445: PetscInt ii;
4446: Mat_SeqAIJ *aij;
4447: #if defined(PETSC_USE_DEBUG)
4448: PetscInt jj;
4449: #endif
4452: if (i[0]) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"i (row indices) must start with 0");
4453: MatCreate(comm,mat);
4454: MatSetSizes(*mat,m,n,m,n);
4455: /* MatSetBlockSizes(*mat,,); */
4456: MatSetType(*mat,MATSEQAIJ);
4457: MatSeqAIJSetPreallocation_SeqAIJ(*mat,MAT_SKIP_ALLOCATION,0);
4458: aij = (Mat_SeqAIJ*)(*mat)->data;
4459: PetscMalloc2(m,&aij->imax,m,&aij->ilen);
4461: aij->i = i;
4462: aij->j = j;
4463: aij->a = a;
4464: aij->singlemalloc = PETSC_FALSE;
4465: aij->nonew = -1; /*this indicates that inserting a new value in the matrix that generates a new nonzero is an error*/
4466: aij->free_a = PETSC_FALSE;
4467: aij->free_ij = PETSC_FALSE;
4469: for (ii=0; ii<m; ii++) {
4470: aij->ilen[ii] = aij->imax[ii] = i[ii+1] - i[ii];
4471: #if defined(PETSC_USE_DEBUG)
4472: 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]);
4473: for (jj=i[ii]+1; jj<i[ii+1]; jj++) {
4474: if (j[jj] < j[jj-1]) SETERRQ3(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column entry number %D (actual colum %D) in row %D is not sorted",jj-i[ii],j[jj],ii);
4475: if (j[jj] == j[jj]-1) SETERRQ3(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column entry number %D (actual colum %D) in row %D is identical to previous entry",jj-i[ii],j[jj],ii);
4476: }
4477: #endif
4478: }
4479: #if defined(PETSC_USE_DEBUG)
4480: for (ii=0; ii<aij->i[m]; ii++) {
4481: if (j[ii] < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative column index at location = %D index = %D",ii,j[ii]);
4482: 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]);
4483: }
4484: #endif
4486: MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);
4487: MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);
4488: return(0);
4489: }
4492: /*@C
4493: MatCreateSeqAIJFromTriple - Creates an sequential AIJ matrix using matrix elements (in COO format)
4494: provided by the user.
4496: Collective on MPI_Comm
4498: Input Parameters:
4499: + comm - must be an MPI communicator of size 1
4500: . m - number of rows
4501: . n - number of columns
4502: . i - row indices
4503: . j - column indices
4504: . a - matrix values
4505: . nz - number of nonzeros
4506: - idx - 0 or 1 based
4508: Output Parameter:
4509: . mat - the matrix
4511: Level: intermediate
4513: Notes:
4514: The i and j indices are 0 based
4516: The format which is used for the sparse matrix input, is equivalent to a
4517: row-major ordering.. i.e for the following matrix, the input data expected is
4518: as shown:
4520: 1 0 0
4521: 2 0 3
4522: 4 5 6
4524: i = {0,1,1,2,2,2}
4525: j = {0,0,2,0,1,2}
4526: v = {1,2,3,4,5,6}
4529: .seealso: MatCreate(), MatCreateAIJ(), MatCreateSeqAIJ(), MatCreateSeqAIJWithArrays(), MatMPIAIJSetPreallocationCSR()
4531: @*/
4532: PetscErrorCode MatCreateSeqAIJFromTriple(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt *i,PetscInt *j,PetscScalar *a,Mat *mat,PetscInt nz,PetscBool idx)
4533: {
4535: PetscInt ii, *nnz, one = 1,row,col;
4539: PetscCalloc1(m,&nnz);
4540: for (ii = 0; ii < nz; ii++) {
4541: nnz[i[ii] - !!idx] += 1;
4542: }
4543: MatCreate(comm,mat);
4544: MatSetSizes(*mat,m,n,m,n);
4545: MatSetType(*mat,MATSEQAIJ);
4546: MatSeqAIJSetPreallocation_SeqAIJ(*mat,0,nnz);
4547: for (ii = 0; ii < nz; ii++) {
4548: if (idx) {
4549: row = i[ii] - 1;
4550: col = j[ii] - 1;
4551: } else {
4552: row = i[ii];
4553: col = j[ii];
4554: }
4555: MatSetValues(*mat,one,&row,one,&col,&a[ii],ADD_VALUES);
4556: }
4557: MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);
4558: MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);
4559: PetscFree(nnz);
4560: return(0);
4561: }
4565: PetscErrorCode MatSetColoring_SeqAIJ(Mat A,ISColoring coloring)
4566: {
4568: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
4571: if (coloring->ctype == IS_COLORING_GLOBAL) {
4572: ISColoringReference(coloring);
4573: a->coloring = coloring;
4574: } else if (coloring->ctype == IS_COLORING_GHOSTED) {
4575: PetscInt i,*larray;
4576: ISColoring ocoloring;
4577: ISColoringValue *colors;
4579: /* set coloring for diagonal portion */
4580: PetscMalloc1(A->cmap->n,&larray);
4581: for (i=0; i<A->cmap->n; i++) larray[i] = i;
4582: ISGlobalToLocalMappingApply(A->cmap->mapping,IS_GTOLM_MASK,A->cmap->n,larray,NULL,larray);
4583: PetscMalloc1(A->cmap->n,&colors);
4584: for (i=0; i<A->cmap->n; i++) colors[i] = coloring->colors[larray[i]];
4585: PetscFree(larray);
4586: ISColoringCreate(PETSC_COMM_SELF,coloring->n,A->cmap->n,colors,&ocoloring);
4587: a->coloring = ocoloring;
4588: }
4589: return(0);
4590: }
4594: PetscErrorCode MatSetValuesAdifor_SeqAIJ(Mat A,PetscInt nl,void *advalues)
4595: {
4596: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
4597: PetscInt m = A->rmap->n,*ii = a->i,*jj = a->j,nz,i,j;
4598: MatScalar *v = a->a;
4599: PetscScalar *values = (PetscScalar*)advalues;
4600: ISColoringValue *color;
4603: if (!a->coloring) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Coloring not set for matrix");
4604: color = a->coloring->colors;
4605: /* loop over rows */
4606: for (i=0; i<m; i++) {
4607: nz = ii[i+1] - ii[i];
4608: /* loop over columns putting computed value into matrix */
4609: for (j=0; j<nz; j++) *v++ = values[color[*jj++]];
4610: values += nl; /* jump to next row of derivatives */
4611: }
4612: return(0);
4613: }
4617: PetscErrorCode MatSeqAIJInvalidateDiagonal(Mat A)
4618: {
4619: Mat_SeqAIJ *a=(Mat_SeqAIJ*)A->data;
4623: a->idiagvalid = PETSC_FALSE;
4624: a->ibdiagvalid = PETSC_FALSE;
4626: MatSeqAIJInvalidateDiagonal_Inode(A);
4627: return(0);
4628: }
4630: /*
4631: Special version for direct calls from Fortran
4632: */
4633: #include <petsc-private/fortranimpl.h>
4634: #if defined(PETSC_HAVE_FORTRAN_CAPS)
4635: #define matsetvaluesseqaij_ MATSETVALUESSEQAIJ
4636: #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE)
4637: #define matsetvaluesseqaij_ matsetvaluesseqaij
4638: #endif
4640: /* Change these macros so can be used in void function */
4641: #undef CHKERRQ
4642: #define CHKERRQ(ierr) CHKERRABORT(PetscObjectComm((PetscObject)A),ierr)
4643: #undef SETERRQ2
4644: #define SETERRQ2(comm,ierr,b,c,d) CHKERRABORT(comm,ierr)
4645: #undef SETERRQ3
4646: #define SETERRQ3(comm,ierr,b,c,d,e) CHKERRABORT(comm,ierr)
4650: PETSC_EXTERN void PETSC_STDCALL matsetvaluesseqaij_(Mat *AA,PetscInt *mm,const PetscInt im[],PetscInt *nn,const PetscInt in[],const PetscScalar v[],InsertMode *isis, PetscErrorCode *_ierr)
4651: {
4652: Mat A = *AA;
4653: PetscInt m = *mm, n = *nn;
4654: InsertMode is = *isis;
4655: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
4656: PetscInt *rp,k,low,high,t,ii,row,nrow,i,col,l,rmax,N;
4657: PetscInt *imax,*ai,*ailen;
4659: PetscInt *aj,nonew = a->nonew,lastcol = -1;
4660: MatScalar *ap,value,*aa;
4661: PetscBool ignorezeroentries = a->ignorezeroentries;
4662: PetscBool roworiented = a->roworiented;
4665: MatCheckPreallocated(A,1);
4666: imax = a->imax;
4667: ai = a->i;
4668: ailen = a->ilen;
4669: aj = a->j;
4670: aa = a->a;
4672: for (k=0; k<m; k++) { /* loop over added rows */
4673: row = im[k];
4674: if (row < 0) continue;
4675: #if defined(PETSC_USE_DEBUG)
4676: if (row >= A->rmap->n) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Row too large");
4677: #endif
4678: rp = aj + ai[row]; ap = aa + ai[row];
4679: rmax = imax[row]; nrow = ailen[row];
4680: low = 0;
4681: high = nrow;
4682: for (l=0; l<n; l++) { /* loop over added columns */
4683: if (in[l] < 0) continue;
4684: #if defined(PETSC_USE_DEBUG)
4685: if (in[l] >= A->cmap->n) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Column too large");
4686: #endif
4687: col = in[l];
4688: if (roworiented) value = v[l + k*n];
4689: else value = v[k + l*m];
4691: if (value == 0.0 && ignorezeroentries && (is == ADD_VALUES)) continue;
4693: if (col <= lastcol) low = 0;
4694: else high = nrow;
4695: lastcol = col;
4696: while (high-low > 5) {
4697: t = (low+high)/2;
4698: if (rp[t] > col) high = t;
4699: else low = t;
4700: }
4701: for (i=low; i<high; i++) {
4702: if (rp[i] > col) break;
4703: if (rp[i] == col) {
4704: if (is == ADD_VALUES) ap[i] += value;
4705: else ap[i] = value;
4706: goto noinsert;
4707: }
4708: }
4709: if (value == 0.0 && ignorezeroentries) goto noinsert;
4710: if (nonew == 1) goto noinsert;
4711: if (nonew == -1) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero in the matrix");
4712: MatSeqXAIJReallocateAIJ(A,A->rmap->n,1,nrow,row,col,rmax,aa,ai,aj,rp,ap,imax,nonew,MatScalar);
4713: N = nrow++ - 1; a->nz++; high++;
4714: /* shift up all the later entries in this row */
4715: for (ii=N; ii>=i; ii--) {
4716: rp[ii+1] = rp[ii];
4717: ap[ii+1] = ap[ii];
4718: }
4719: rp[i] = col;
4720: ap[i] = value;
4721: A->nonzerostate++;
4722: noinsert:;
4723: low = i + 1;
4724: }
4725: ailen[row] = nrow;
4726: }
4727: PetscFunctionReturnVoid();
4728: }