Actual source code: aij.c
1: #define PETSCMAT_DLL
3: /*
4: Defines the basic matrix operations for the AIJ (compressed row)
5: matrix storage format.
6: */
8: #include src/mat/impls/aij/seq/aij.h
9: #include src/inline/spops.h
10: #include src/inline/dot.h
11: #include petscbt.h
15: PetscErrorCode MatDiagonalSet_SeqAIJ(Mat Y,Vec D,InsertMode is)
16: {
18: Mat_SeqAIJ *aij = (Mat_SeqAIJ*) Y->data;
19: PetscInt i,*diag, m = Y->rmap.n;
20: PetscScalar *v,*aa = aij->a;
21: PetscTruth missing;
24: if (Y->assembled) {
25: MatMissingDiagonal_SeqAIJ(Y,&missing,PETSC_NULL);
26: if (!missing) {
27: diag = aij->diag;
28: VecGetArray(D,&v);
29: if (is == INSERT_VALUES) {
30: for (i=0; i<m; i++) {
31: aa[diag[i]] = v[i];
32: }
33: } else {
34: for (i=0; i<m; i++) {
35: aa[diag[i]] += v[i];
36: }
37: }
38: VecRestoreArray(D,&v);
39: return(0);
40: }
41: }
42: MatDiagonalSet_Default(Y,D,is);
43: return(0);
44: }
48: PetscErrorCode MatGetRowIJ_SeqAIJ(Mat A,PetscInt oshift,PetscTruth symmetric,PetscInt *m,PetscInt *ia[],PetscInt *ja[],PetscTruth *done)
49: {
50: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
52: PetscInt i,ishift;
53:
55: *m = A->rmap.n;
56: if (!ia) return(0);
57: ishift = 0;
58: if (symmetric && !A->structurally_symmetric) {
59: MatToSymmetricIJ_SeqAIJ(A->rmap.n,a->i,a->j,ishift,oshift,ia,ja);
60: } else if (oshift == 1) {
61: PetscInt nz = a->i[A->rmap.n];
62: /* malloc space and add 1 to i and j indices */
63: PetscMalloc((A->rmap.n+1)*sizeof(PetscInt),ia);
64: PetscMalloc((nz+1)*sizeof(PetscInt),ja);
65: for (i=0; i<nz; i++) (*ja)[i] = a->j[i] + 1;
66: for (i=0; i<A->rmap.n+1; i++) (*ia)[i] = a->i[i] + 1;
67: } else {
68: *ia = a->i; *ja = a->j;
69: }
70: return(0);
71: }
75: PetscErrorCode MatRestoreRowIJ_SeqAIJ(Mat A,PetscInt oshift,PetscTruth symmetric,PetscInt *n,PetscInt *ia[],PetscInt *ja[],PetscTruth *done)
76: {
78:
80: if (!ia) return(0);
81: if ((symmetric && !A->structurally_symmetric) || oshift == 1) {
82: PetscFree(*ia);
83: PetscFree(*ja);
84: }
85: return(0);
86: }
90: PetscErrorCode MatGetColumnIJ_SeqAIJ(Mat A,PetscInt oshift,PetscTruth symmetric,PetscInt *nn,PetscInt *ia[],PetscInt *ja[],PetscTruth *done)
91: {
92: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
94: PetscInt i,*collengths,*cia,*cja,n = A->cmap.n,m = A->rmap.n;
95: PetscInt nz = a->i[m],row,*jj,mr,col;
96:
98: *nn = n;
99: if (!ia) return(0);
100: if (symmetric) {
101: MatToSymmetricIJ_SeqAIJ(A->rmap.n,a->i,a->j,0,oshift,ia,ja);
102: } else {
103: PetscMalloc((n+1)*sizeof(PetscInt),&collengths);
104: PetscMemzero(collengths,n*sizeof(PetscInt));
105: PetscMalloc((n+1)*sizeof(PetscInt),&cia);
106: PetscMalloc((nz+1)*sizeof(PetscInt),&cja);
107: jj = a->j;
108: for (i=0; i<nz; i++) {
109: collengths[jj[i]]++;
110: }
111: cia[0] = oshift;
112: for (i=0; i<n; i++) {
113: cia[i+1] = cia[i] + collengths[i];
114: }
115: PetscMemzero(collengths,n*sizeof(PetscInt));
116: jj = a->j;
117: for (row=0; row<m; row++) {
118: mr = a->i[row+1] - a->i[row];
119: for (i=0; i<mr; i++) {
120: col = *jj++;
121: cja[cia[col] + collengths[col]++ - oshift] = row + oshift;
122: }
123: }
124: PetscFree(collengths);
125: *ia = cia; *ja = cja;
126: }
127: return(0);
128: }
132: PetscErrorCode MatRestoreColumnIJ_SeqAIJ(Mat A,PetscInt oshift,PetscTruth symmetric,PetscInt *n,PetscInt *ia[],PetscInt *ja[],PetscTruth *done)
133: {
137: if (!ia) return(0);
139: PetscFree(*ia);
140: PetscFree(*ja);
141:
142: return(0);
143: }
147: PetscErrorCode MatSetValuesRow_SeqAIJ(Mat A,PetscInt row,const PetscScalar v[])
148: {
149: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
150: PetscInt *ai = a->i;
154: PetscMemcpy(a->a+ai[row],v,(ai[row+1]-ai[row])*sizeof(PetscScalar));
155: return(0);
156: }
158: #define CHUNKSIZE 15
162: PetscErrorCode MatSetValues_SeqAIJ(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode is)
163: {
164: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
165: PetscInt *rp,k,low,high,t,ii,row,nrow,i,col,l,rmax,N;
166: PetscInt *imax = a->imax,*ai = a->i,*ailen = a->ilen;
168: PetscInt *aj = a->j,nonew = a->nonew,lastcol = -1;
169: PetscScalar *ap,value,*aa = a->a;
170: PetscTruth ignorezeroentries = ((a->ignorezeroentries && is == ADD_VALUES) ? PETSC_TRUE:PETSC_FALSE);
171: PetscTruth roworiented = a->roworiented;
174: for (k=0; k<m; k++) { /* loop over added rows */
175: row = im[k];
176: if (row < 0) continue;
177: #if defined(PETSC_USE_DEBUG)
178: if (row >= A->rmap.n) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",row,A->rmap.n-1);
179: #endif
180: rp = aj + ai[row]; ap = aa + ai[row];
181: rmax = imax[row]; nrow = ailen[row];
182: low = 0;
183: high = nrow;
184: for (l=0; l<n; l++) { /* loop over added columns */
185: if (in[l] < 0) continue;
186: #if defined(PETSC_USE_DEBUG)
187: if (in[l] >= A->cmap.n) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",in[l],A->cmap.n-1);
188: #endif
189: col = in[l];
190: if (roworiented) {
191: value = v[l + k*n];
192: } else {
193: value = v[k + l*m];
194: }
195: if (value == 0.0 && ignorezeroentries && (is == ADD_VALUES)) continue;
197: if (col <= lastcol) low = 0; else high = nrow;
198: lastcol = col;
199: while (high-low > 5) {
200: t = (low+high)/2;
201: if (rp[t] > col) high = t;
202: else low = t;
203: }
204: for (i=low; i<high; i++) {
205: if (rp[i] > col) break;
206: if (rp[i] == col) {
207: if (is == ADD_VALUES) ap[i] += value;
208: else ap[i] = value;
209: goto noinsert;
210: }
211: }
212: if (value == 0.0 && ignorezeroentries) goto noinsert;
213: if (nonew == 1) goto noinsert;
214: if (nonew == -1) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero at (%D,%D) in the matrix",row,col);
215: MatSeqXAIJReallocateAIJ(A,A->rmap.n,1,nrow,row,col,rmax,aa,ai,aj,rp,ap,imax,nonew,MatScalar);
216: N = nrow++ - 1; a->nz++; high++;
217: /* shift up all the later entries in this row */
218: for (ii=N; ii>=i; ii--) {
219: rp[ii+1] = rp[ii];
220: ap[ii+1] = ap[ii];
221: }
222: rp[i] = col;
223: ap[i] = value;
224: noinsert:;
225: low = i + 1;
226: }
227: ailen[row] = nrow;
228: }
229: A->same_nonzero = PETSC_FALSE;
230: return(0);
231: }
236: PetscErrorCode MatGetValues_SeqAIJ(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],PetscScalar v[])
237: {
238: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
239: PetscInt *rp,k,low,high,t,row,nrow,i,col,l,*aj = a->j;
240: PetscInt *ai = a->i,*ailen = a->ilen;
241: PetscScalar *ap,*aa = a->a,zero = 0.0;
244: for (k=0; k<m; k++) { /* loop over rows */
245: row = im[k];
246: if (row < 0) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"Negative row: %D",row);
247: if (row >= A->rmap.n) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",row,A->rmap.n-1);
248: rp = aj + ai[row]; ap = aa + ai[row];
249: nrow = ailen[row];
250: for (l=0; l<n; l++) { /* loop over columns */
251: if (in[l] < 0) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"Negative column: %D",in[l]);
252: if (in[l] >= A->cmap.n) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",in[l],A->cmap.n-1);
253: col = in[l] ;
254: high = nrow; low = 0; /* assume unsorted */
255: while (high-low > 5) {
256: t = (low+high)/2;
257: if (rp[t] > col) high = t;
258: else low = t;
259: }
260: for (i=low; i<high; i++) {
261: if (rp[i] > col) break;
262: if (rp[i] == col) {
263: *v++ = ap[i];
264: goto finished;
265: }
266: }
267: *v++ = zero;
268: finished:;
269: }
270: }
271: return(0);
272: }
277: PetscErrorCode MatView_SeqAIJ_Binary(Mat A,PetscViewer viewer)
278: {
279: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
281: PetscInt i,*col_lens;
282: int fd;
285: PetscViewerBinaryGetDescriptor(viewer,&fd);
286: PetscMalloc((4+A->rmap.n)*sizeof(PetscInt),&col_lens);
287: col_lens[0] = MAT_FILE_COOKIE;
288: col_lens[1] = A->rmap.n;
289: col_lens[2] = A->cmap.n;
290: col_lens[3] = a->nz;
292: /* store lengths of each row and write (including header) to file */
293: for (i=0; i<A->rmap.n; i++) {
294: col_lens[4+i] = a->i[i+1] - a->i[i];
295: }
296: PetscBinaryWrite(fd,col_lens,4+A->rmap.n,PETSC_INT,PETSC_TRUE);
297: PetscFree(col_lens);
299: /* store column indices (zero start index) */
300: PetscBinaryWrite(fd,a->j,a->nz,PETSC_INT,PETSC_FALSE);
302: /* store nonzero values */
303: PetscBinaryWrite(fd,a->a,a->nz,PETSC_SCALAR,PETSC_FALSE);
304: return(0);
305: }
307: EXTERN PetscErrorCode MatSeqAIJFactorInfo_Matlab(Mat,PetscViewer);
311: PetscErrorCode MatView_SeqAIJ_ASCII(Mat A,PetscViewer viewer)
312: {
313: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
314: PetscErrorCode ierr;
315: PetscInt i,j,m = A->rmap.n,shift=0;
316: const char *name;
317: PetscViewerFormat format;
320: PetscObjectGetName((PetscObject)A,&name);
321: PetscViewerGetFormat(viewer,&format);
322: if (format == PETSC_VIEWER_ASCII_MATLAB) {
323: PetscInt nofinalvalue = 0;
324: if ((a->i[m] == a->i[m-1]) || (a->j[a->nz-1] != A->cmap.n-!shift)) {
325: nofinalvalue = 1;
326: }
327: PetscViewerASCIIUseTabs(viewer,PETSC_NO);
328: PetscViewerASCIIPrintf(viewer,"%% Size = %D %D \n",m,A->cmap.n);
329: PetscViewerASCIIPrintf(viewer,"%% Nonzeros = %D \n",a->nz);
330: PetscViewerASCIIPrintf(viewer,"zzz = zeros(%D,3);\n",a->nz+nofinalvalue);
331: PetscViewerASCIIPrintf(viewer,"zzz = [\n");
333: for (i=0; i<m; i++) {
334: for (j=a->i[i]+shift; j<a->i[i+1]+shift; j++) {
335: #if defined(PETSC_USE_COMPLEX)
336: PetscViewerASCIIPrintf(viewer,"%D %D %18.16e + %18.16ei \n",i+1,a->j[j]+!shift,PetscRealPart(a->a[j]),PetscImaginaryPart(a->a[j]));
337: #else
338: PetscViewerASCIIPrintf(viewer,"%D %D %18.16e\n",i+1,a->j[j]+!shift,a->a[j]);
339: #endif
340: }
341: }
342: if (nofinalvalue) {
343: PetscViewerASCIIPrintf(viewer,"%D %D %18.16e\n",m,A->cmap.n,0.0);
344: }
345: PetscViewerASCIIPrintf(viewer,"];\n %s = spconvert(zzz);\n",name);
346: PetscViewerASCIIUseTabs(viewer,PETSC_YES);
347: } else if (format == PETSC_VIEWER_ASCII_FACTOR_INFO || format == PETSC_VIEWER_ASCII_INFO) {
348: return(0);
349: } else if (format == PETSC_VIEWER_ASCII_COMMON) {
350: PetscViewerASCIIUseTabs(viewer,PETSC_NO);
351: for (i=0; i<m; i++) {
352: PetscViewerASCIIPrintf(viewer,"row %D:",i);
353: for (j=a->i[i]+shift; j<a->i[i+1]+shift; j++) {
354: #if defined(PETSC_USE_COMPLEX)
355: if (PetscImaginaryPart(a->a[j]) > 0.0 && PetscRealPart(a->a[j]) != 0.0) {
356: PetscViewerASCIIPrintf(viewer," (%D, %G + %G i)",a->j[j]+shift,PetscRealPart(a->a[j]),PetscImaginaryPart(a->a[j]));
357: } else if (PetscImaginaryPart(a->a[j]) < 0.0 && PetscRealPart(a->a[j]) != 0.0) {
358: PetscViewerASCIIPrintf(viewer," (%D, %G - %G i)",a->j[j]+shift,PetscRealPart(a->a[j]),-PetscImaginaryPart(a->a[j]));
359: } else if (PetscRealPart(a->a[j]) != 0.0) {
360: PetscViewerASCIIPrintf(viewer," (%D, %G) ",a->j[j]+shift,PetscRealPart(a->a[j]));
361: }
362: #else
363: if (a->a[j] != 0.0) {PetscViewerASCIIPrintf(viewer," (%D, %G) ",a->j[j]+shift,a->a[j]);}
364: #endif
365: }
366: PetscViewerASCIIPrintf(viewer,"\n");
367: }
368: PetscViewerASCIIUseTabs(viewer,PETSC_YES);
369: } else if (format == PETSC_VIEWER_ASCII_SYMMODU) {
370: PetscInt nzd=0,fshift=1,*sptr;
371: PetscViewerASCIIUseTabs(viewer,PETSC_NO);
372: PetscMalloc((m+1)*sizeof(PetscInt),&sptr);
373: for (i=0; i<m; i++) {
374: sptr[i] = nzd+1;
375: for (j=a->i[i]+shift; j<a->i[i+1]+shift; j++) {
376: if (a->j[j] >= i) {
377: #if defined(PETSC_USE_COMPLEX)
378: if (PetscImaginaryPart(a->a[j]) != 0.0 || PetscRealPart(a->a[j]) != 0.0) nzd++;
379: #else
380: if (a->a[j] != 0.0) nzd++;
381: #endif
382: }
383: }
384: }
385: sptr[m] = nzd+1;
386: PetscViewerASCIIPrintf(viewer," %D %D\n\n",m,nzd);
387: for (i=0; i<m+1; i+=6) {
388: if (i+4<m) {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]);}
389: else if (i+3<m) {PetscViewerASCIIPrintf(viewer," %D %D %D %D %D\n",sptr[i],sptr[i+1],sptr[i+2],sptr[i+3],sptr[i+4]);}
390: else if (i+2<m) {PetscViewerASCIIPrintf(viewer," %D %D %D %D\n",sptr[i],sptr[i+1],sptr[i+2],sptr[i+3]);}
391: else if (i+1<m) {PetscViewerASCIIPrintf(viewer," %D %D %D\n",sptr[i],sptr[i+1],sptr[i+2]);}
392: else if (i<m) {PetscViewerASCIIPrintf(viewer," %D %D\n",sptr[i],sptr[i+1]);}
393: else {PetscViewerASCIIPrintf(viewer," %D\n",sptr[i]);}
394: }
395: PetscViewerASCIIPrintf(viewer,"\n");
396: PetscFree(sptr);
397: for (i=0; i<m; i++) {
398: for (j=a->i[i]+shift; j<a->i[i+1]+shift; j++) {
399: if (a->j[j] >= i) {PetscViewerASCIIPrintf(viewer," %D ",a->j[j]+fshift);}
400: }
401: PetscViewerASCIIPrintf(viewer,"\n");
402: }
403: PetscViewerASCIIPrintf(viewer,"\n");
404: for (i=0; i<m; i++) {
405: for (j=a->i[i]+shift; j<a->i[i+1]+shift; j++) {
406: if (a->j[j] >= i) {
407: #if defined(PETSC_USE_COMPLEX)
408: if (PetscImaginaryPart(a->a[j]) != 0.0 || PetscRealPart(a->a[j]) != 0.0) {
409: PetscViewerASCIIPrintf(viewer," %18.16e %18.16e ",PetscRealPart(a->a[j]),PetscImaginaryPart(a->a[j]));
410: }
411: #else
412: if (a->a[j] != 0.0) {PetscViewerASCIIPrintf(viewer," %18.16e ",a->a[j]);}
413: #endif
414: }
415: }
416: PetscViewerASCIIPrintf(viewer,"\n");
417: }
418: PetscViewerASCIIUseTabs(viewer,PETSC_YES);
419: } else if (format == PETSC_VIEWER_ASCII_DENSE) {
420: PetscInt cnt = 0,jcnt;
421: PetscScalar value;
423: PetscViewerASCIIUseTabs(viewer,PETSC_NO);
424: for (i=0; i<m; i++) {
425: jcnt = 0;
426: for (j=0; j<A->cmap.n; j++) {
427: if (jcnt < a->i[i+1]-a->i[i] && j == a->j[cnt]) {
428: value = a->a[cnt++];
429: jcnt++;
430: } else {
431: value = 0.0;
432: }
433: #if defined(PETSC_USE_COMPLEX)
434: PetscViewerASCIIPrintf(viewer," %7.5e+%7.5e i ",PetscRealPart(value),PetscImaginaryPart(value));
435: #else
436: PetscViewerASCIIPrintf(viewer," %7.5e ",value);
437: #endif
438: }
439: PetscViewerASCIIPrintf(viewer,"\n");
440: }
441: PetscViewerASCIIUseTabs(viewer,PETSC_YES);
442: } else {
443: PetscViewerASCIIUseTabs(viewer,PETSC_NO);
444: for (i=0; i<m; i++) {
445: PetscViewerASCIIPrintf(viewer,"row %D:",i);
446: for (j=a->i[i]+shift; j<a->i[i+1]+shift; j++) {
447: #if defined(PETSC_USE_COMPLEX)
448: if (PetscImaginaryPart(a->a[j]) > 0.0) {
449: PetscViewerASCIIPrintf(viewer," (%D, %G + %G i)",a->j[j]+shift,PetscRealPart(a->a[j]),PetscImaginaryPart(a->a[j]));
450: } else if (PetscImaginaryPart(a->a[j]) < 0.0) {
451: PetscViewerASCIIPrintf(viewer," (%D, %G - %G i)",a->j[j]+shift,PetscRealPart(a->a[j]),-PetscImaginaryPart(a->a[j]));
452: } else {
453: PetscViewerASCIIPrintf(viewer," (%D, %G) ",a->j[j]+shift,PetscRealPart(a->a[j]));
454: }
455: #else
456: PetscViewerASCIIPrintf(viewer," (%D, %G) ",a->j[j]+shift,a->a[j]);
457: #endif
458: }
459: PetscViewerASCIIPrintf(viewer,"\n");
460: }
461: PetscViewerASCIIUseTabs(viewer,PETSC_YES);
462: }
463: PetscViewerFlush(viewer);
464: return(0);
465: }
469: PetscErrorCode MatView_SeqAIJ_Draw_Zoom(PetscDraw draw,void *Aa)
470: {
471: Mat A = (Mat) Aa;
472: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
473: PetscErrorCode ierr;
474: PetscInt i,j,m = A->rmap.n,color;
475: PetscReal xl,yl,xr,yr,x_l,x_r,y_l,y_r,maxv = 0.0;
476: PetscViewer viewer;
477: PetscViewerFormat format;
480: PetscObjectQuery((PetscObject)A,"Zoomviewer",(PetscObject*)&viewer);
481: PetscViewerGetFormat(viewer,&format);
483: PetscDrawGetCoordinates(draw,&xl,&yl,&xr,&yr);
484: /* loop over matrix elements drawing boxes */
486: if (format != PETSC_VIEWER_DRAW_CONTOUR) {
487: /* Blue for negative, Cyan for zero and Red for positive */
488: color = PETSC_DRAW_BLUE;
489: for (i=0; i<m; i++) {
490: y_l = m - i - 1.0; y_r = y_l + 1.0;
491: for (j=a->i[i]; j<a->i[i+1]; j++) {
492: x_l = a->j[j] ; x_r = x_l + 1.0;
493: #if defined(PETSC_USE_COMPLEX)
494: if (PetscRealPart(a->a[j]) >= 0.) continue;
495: #else
496: if (a->a[j] >= 0.) continue;
497: #endif
498: PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);
499: }
500: }
501: color = PETSC_DRAW_CYAN;
502: for (i=0; i<m; i++) {
503: y_l = m - i - 1.0; y_r = y_l + 1.0;
504: for (j=a->i[i]; j<a->i[i+1]; j++) {
505: x_l = a->j[j]; x_r = x_l + 1.0;
506: if (a->a[j] != 0.) continue;
507: PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);
508: }
509: }
510: color = PETSC_DRAW_RED;
511: for (i=0; i<m; i++) {
512: y_l = m - i - 1.0; y_r = y_l + 1.0;
513: for (j=a->i[i]; j<a->i[i+1]; j++) {
514: x_l = a->j[j]; x_r = x_l + 1.0;
515: #if defined(PETSC_USE_COMPLEX)
516: if (PetscRealPart(a->a[j]) <= 0.) continue;
517: #else
518: if (a->a[j] <= 0.) continue;
519: #endif
520: PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);
521: }
522: }
523: } else {
524: /* use contour shading to indicate magnitude of values */
525: /* first determine max of all nonzero values */
526: PetscInt nz = a->nz,count;
527: PetscDraw popup;
528: PetscReal scale;
530: for (i=0; i<nz; i++) {
531: if (PetscAbsScalar(a->a[i]) > maxv) maxv = PetscAbsScalar(a->a[i]);
532: }
533: scale = (245.0 - PETSC_DRAW_BASIC_COLORS)/maxv;
534: PetscDrawGetPopup(draw,&popup);
535: if (popup) {PetscDrawScalePopup(popup,0.0,maxv);}
536: count = 0;
537: for (i=0; i<m; i++) {
538: y_l = m - i - 1.0; y_r = y_l + 1.0;
539: for (j=a->i[i]; j<a->i[i+1]; j++) {
540: x_l = a->j[j]; x_r = x_l + 1.0;
541: color = PETSC_DRAW_BASIC_COLORS + (PetscInt)(scale*PetscAbsScalar(a->a[count]));
542: PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);
543: count++;
544: }
545: }
546: }
547: return(0);
548: }
552: PetscErrorCode MatView_SeqAIJ_Draw(Mat A,PetscViewer viewer)
553: {
555: PetscDraw draw;
556: PetscReal xr,yr,xl,yl,h,w;
557: PetscTruth isnull;
560: PetscViewerDrawGetDraw(viewer,0,&draw);
561: PetscDrawIsNull(draw,&isnull);
562: if (isnull) return(0);
564: PetscObjectCompose((PetscObject)A,"Zoomviewer",(PetscObject)viewer);
565: xr = A->cmap.n; yr = A->rmap.n; h = yr/10.0; w = xr/10.0;
566: xr += w; yr += h; xl = -w; yl = -h;
567: PetscDrawSetCoordinates(draw,xl,yl,xr,yr);
568: PetscDrawZoom(draw,MatView_SeqAIJ_Draw_Zoom,A);
569: PetscObjectCompose((PetscObject)A,"Zoomviewer",PETSC_NULL);
570: return(0);
571: }
575: PetscErrorCode MatView_SeqAIJ(Mat A,PetscViewer viewer)
576: {
578: PetscTruth iascii,isbinary,isdraw;
581: PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_ASCII,&iascii);
582: PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_BINARY,&isbinary);
583: PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_DRAW,&isdraw);
584: if (iascii) {
585: MatView_SeqAIJ_ASCII(A,viewer);
586: } else if (isbinary) {
587: MatView_SeqAIJ_Binary(A,viewer);
588: } else if (isdraw) {
589: MatView_SeqAIJ_Draw(A,viewer);
590: } else {
591: SETERRQ1(PETSC_ERR_SUP,"Viewer type %s not supported by SeqAIJ matrices",((PetscObject)viewer)->type_name);
592: }
593: MatView_Inode(A,viewer);
594: return(0);
595: }
599: PetscErrorCode MatAssemblyEnd_SeqAIJ(Mat A,MatAssemblyType mode)
600: {
601: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
603: PetscInt fshift = 0,i,j,*ai = a->i,*aj = a->j,*imax = a->imax;
604: PetscInt m = A->rmap.n,*ip,N,*ailen = a->ilen,rmax = 0;
605: PetscScalar *aa = a->a,*ap;
606: PetscReal ratio=0.6;
609: if (mode == MAT_FLUSH_ASSEMBLY) return(0);
611: if (m) rmax = ailen[0]; /* determine row with most nonzeros */
612: for (i=1; i<m; i++) {
613: /* move each row back by the amount of empty slots (fshift) before it*/
614: fshift += imax[i-1] - ailen[i-1];
615: rmax = PetscMax(rmax,ailen[i]);
616: if (fshift) {
617: ip = aj + ai[i] ;
618: ap = aa + ai[i] ;
619: N = ailen[i];
620: for (j=0; j<N; j++) {
621: ip[j-fshift] = ip[j];
622: ap[j-fshift] = ap[j];
623: }
624: }
625: ai[i] = ai[i-1] + ailen[i-1];
626: }
627: if (m) {
628: fshift += imax[m-1] - ailen[m-1];
629: ai[m] = ai[m-1] + ailen[m-1];
630: }
631: /* reset ilen and imax for each row */
632: for (i=0; i<m; i++) {
633: ailen[i] = imax[i] = ai[i+1] - ai[i];
634: }
635: a->nz = ai[m];
637: MatMarkDiagonal_SeqAIJ(A);
638: PetscInfo4(A,"Matrix size: %D X %D; storage space: %D unneeded,%D used\n",m,A->cmap.n,fshift,a->nz);
639: PetscInfo1(A,"Number of mallocs during MatSetValues() is %D\n",a->reallocs);
640: PetscInfo1(A,"Maximum nonzeros in any row is %D\n",rmax);
642: a->reallocs = 0;
643: A->info.nz_unneeded = (double)fshift;
644: a->rmax = rmax;
646: /* check for zero rows. If found a large number of zero rows, use CompressedRow functions */
647: Mat_CheckCompressedRow(A,&a->compressedrow,a->i,m,ratio);
648: A->same_nonzero = PETSC_TRUE;
650: MatAssemblyEnd_Inode(A,mode);
651: return(0);
652: }
656: PetscErrorCode MatRealPart_SeqAIJ(Mat A)
657: {
658: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
659: PetscInt i,nz = a->nz;
660: PetscScalar *aa = a->a;
663: for (i=0; i<nz; i++) aa[i] = PetscRealPart(aa[i]);
664: return(0);
665: }
669: PetscErrorCode MatImaginaryPart_SeqAIJ(Mat A)
670: {
671: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
672: PetscInt i,nz = a->nz;
673: PetscScalar *aa = a->a;
676: for (i=0; i<nz; i++) aa[i] = PetscImaginaryPart(aa[i]);
677: return(0);
678: }
682: PetscErrorCode MatZeroEntries_SeqAIJ(Mat A)
683: {
684: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
688: PetscMemzero(a->a,(a->i[A->rmap.n])*sizeof(PetscScalar));
689: return(0);
690: }
694: PetscErrorCode MatDestroy_SeqAIJ(Mat A)
695: {
696: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
700: #if defined(PETSC_USE_LOG)
701: PetscLogObjectState((PetscObject)A,"Rows=%D, Cols=%D, NZ=%D",A->rmap.n,A->cmap.n,a->nz);
702: #endif
703: MatSeqXAIJFreeAIJ(A,&a->a,&a->j,&a->i);
704: if (a->row) {
705: ISDestroy(a->row);
706: }
707: if (a->col) {
708: ISDestroy(a->col);
709: }
710: PetscFree(a->diag);
711: PetscFree2(a->imax,a->ilen);
712: PetscFree(a->idiag);
713: PetscFree(a->solve_work);
714: if (a->icol) {ISDestroy(a->icol);}
715: PetscFree(a->saved_values);
716: if (a->coloring) {ISColoringDestroy(a->coloring);}
717: PetscFree(a->xtoy);
718: if (a->compressedrow.use){PetscFree(a->compressedrow.i);}
720: MatDestroy_Inode(A);
722: PetscFree(a);
724: PetscObjectChangeTypeName((PetscObject)A,0);
725: PetscObjectComposeFunctionDynamic((PetscObject)A,"MatSeqAIJSetColumnIndices_C","",PETSC_NULL);
726: PetscObjectComposeFunctionDynamic((PetscObject)A,"MatStoreValues_C","",PETSC_NULL);
727: PetscObjectComposeFunctionDynamic((PetscObject)A,"MatRetrieveValues_C","",PETSC_NULL);
728: PetscObjectComposeFunctionDynamic((PetscObject)A,"MatConvert_seqaij_seqsbaij_C","",PETSC_NULL);
729: PetscObjectComposeFunctionDynamic((PetscObject)A,"MatConvert_seqaij_seqbaij_C","",PETSC_NULL);
730: PetscObjectComposeFunctionDynamic((PetscObject)A,"MatConvert_seqaij_seqcsrperm_C","",PETSC_NULL);
731: PetscObjectComposeFunctionDynamic((PetscObject)A,"MatIsTranspose_C","",PETSC_NULL);
732: PetscObjectComposeFunctionDynamic((PetscObject)A,"MatSeqAIJSetPreallocation_C","",PETSC_NULL);
733: PetscObjectComposeFunctionDynamic((PetscObject)A,"MatSeqAIJSetPreallocationCSR_C","",PETSC_NULL);
734: PetscObjectComposeFunctionDynamic((PetscObject)A,"MatReorderForNonzeroDiagonal_C","",PETSC_NULL);
735: return(0);
736: }
740: PetscErrorCode MatCompress_SeqAIJ(Mat A)
741: {
743: return(0);
744: }
748: PetscErrorCode MatSetOption_SeqAIJ(Mat A,MatOption op)
749: {
750: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
754: switch (op) {
755: case MAT_ROW_ORIENTED:
756: a->roworiented = PETSC_TRUE;
757: break;
758: case MAT_KEEP_ZEROED_ROWS:
759: a->keepzeroedrows = PETSC_TRUE;
760: break;
761: case MAT_COLUMN_ORIENTED:
762: a->roworiented = PETSC_FALSE;
763: break;
764: case MAT_COLUMNS_SORTED:
765: a->sorted = PETSC_TRUE;
766: break;
767: case MAT_COLUMNS_UNSORTED:
768: a->sorted = PETSC_FALSE;
769: break;
770: case MAT_NO_NEW_NONZERO_LOCATIONS:
771: a->nonew = 1;
772: break;
773: case MAT_NEW_NONZERO_LOCATION_ERR:
774: a->nonew = -1;
775: break;
776: case MAT_NEW_NONZERO_ALLOCATION_ERR:
777: a->nonew = -2;
778: break;
779: case MAT_YES_NEW_NONZERO_LOCATIONS:
780: a->nonew = 0;
781: break;
782: case MAT_IGNORE_ZERO_ENTRIES:
783: a->ignorezeroentries = PETSC_TRUE;
784: break;
785: case MAT_USE_COMPRESSEDROW:
786: a->compressedrow.use = PETSC_TRUE;
787: break;
788: case MAT_DO_NOT_USE_COMPRESSEDROW:
789: a->compressedrow.use = PETSC_FALSE;
790: break;
791: case MAT_ROWS_SORTED:
792: case MAT_ROWS_UNSORTED:
793: case MAT_YES_NEW_DIAGONALS:
794: case MAT_IGNORE_OFF_PROC_ENTRIES:
795: case MAT_USE_HASH_TABLE:
796: PetscInfo1(A,"Option %s ignored\n",MatOptions[op]);
797: break;
798: case MAT_NO_NEW_DIAGONALS:
799: SETERRQ(PETSC_ERR_SUP,"MAT_NO_NEW_DIAGONALS");
800: default:
801: break;
802: }
803: MatSetOption_Inode(A,op);
804: return(0);
805: }
809: PetscErrorCode MatGetDiagonal_SeqAIJ(Mat A,Vec v)
810: {
811: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
813: PetscInt i,j,n;
814: PetscScalar *x,zero = 0.0;
817: VecSet(v,zero);
818: VecGetArray(v,&x);
819: VecGetLocalSize(v,&n);
820: if (n != A->rmap.n) SETERRQ(PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector");
821: for (i=0; i<A->rmap.n; i++) {
822: for (j=a->i[i]; j<a->i[i+1]; j++) {
823: if (a->j[j] == i) {
824: x[i] = a->a[j];
825: break;
826: }
827: }
828: }
829: VecRestoreArray(v,&x);
830: return(0);
831: }
835: PetscErrorCode MatMultTransposeAdd_SeqAIJ(Mat A,Vec xx,Vec zz,Vec yy)
836: {
837: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
838: PetscScalar *x,*y;
839: PetscErrorCode ierr;
840: PetscInt m = A->rmap.n;
841: #if !defined(PETSC_USE_FORTRAN_KERNEL_MULTTRANSPOSEAIJ)
842: PetscScalar *v,alpha;
843: PetscInt n,i,*idx,*ii,*ridx=PETSC_NULL;
844: Mat_CompressedRow cprow = a->compressedrow;
845: PetscTruth usecprow = cprow.use;
846: #endif
849: if (zz != yy) {VecCopy(zz,yy);}
850: VecGetArray(xx,&x);
851: VecGetArray(yy,&y);
853: #if defined(PETSC_USE_FORTRAN_KERNEL_MULTTRANSPOSEAIJ)
854: fortranmulttransposeaddaij_(&m,x,a->i,a->j,a->a,y);
855: #else
856: if (usecprow){
857: m = cprow.nrows;
858: ii = cprow.i;
859: ridx = cprow.rindex;
860: } else {
861: ii = a->i;
862: }
863: for (i=0; i<m; i++) {
864: idx = a->j + ii[i] ;
865: v = a->a + ii[i] ;
866: n = ii[i+1] - ii[i];
867: if (usecprow){
868: alpha = x[ridx[i]];
869: } else {
870: alpha = x[i];
871: }
872: while (n-->0) {y[*idx++] += alpha * *v++;}
873: }
874: #endif
875: PetscLogFlops(2*a->nz);
876: VecRestoreArray(xx,&x);
877: VecRestoreArray(yy,&y);
878: return(0);
879: }
883: PetscErrorCode MatMultTranspose_SeqAIJ(Mat A,Vec xx,Vec yy)
884: {
885: PetscScalar zero = 0.0;
889: VecSet(yy,zero);
890: MatMultTransposeAdd_SeqAIJ(A,xx,yy,yy);
891: return(0);
892: }
897: PetscErrorCode MatMult_SeqAIJ(Mat A,Vec xx,Vec yy)
898: {
899: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
900: PetscScalar *x,*y,*aa;
902: PetscInt m=A->rmap.n,*aj,*ii;
903: PetscInt n,i,j,*ridx=PETSC_NULL;
904: PetscScalar sum;
905: PetscTruth usecprow=a->compressedrow.use;
906: #if !defined(PETSC_USE_FORTRAN_KERNEL_MULTAIJ)
907: PetscInt jrow;
908: #endif
910: #if defined(PETSC_HAVE_PRAGMA_DISJOINT)
911: #pragma disjoint(*x,*y,*aa)
912: #endif
915: VecGetArray(xx,&x);
916: VecGetArray(yy,&y);
917: aj = a->j;
918: aa = a->a;
919: ii = a->i;
920: if (usecprow){ /* use compressed row format */
921: m = a->compressedrow.nrows;
922: ii = a->compressedrow.i;
923: ridx = a->compressedrow.rindex;
924: for (i=0; i<m; i++){
925: n = ii[i+1] - ii[i];
926: aj = a->j + ii[i];
927: aa = a->a + ii[i];
928: sum = 0.0;
929: for (j=0; j<n; j++) sum += (*aa++)*x[*aj++];
930: y[*ridx++] = sum;
931: }
932: } else { /* do not use compressed row format */
933: #if defined(PETSC_USE_FORTRAN_KERNEL_MULTAIJ)
934: fortranmultaij_(&m,x,ii,aj,aa,y);
935: #else
936: for (i=0; i<m; i++) {
937: jrow = ii[i];
938: n = ii[i+1] - jrow;
939: sum = 0.0;
940: for (j=0; j<n; j++) {
941: sum += aa[jrow]*x[aj[jrow]]; jrow++;
942: }
943: y[i] = sum;
944: }
945: #endif
946: }
947: PetscLogFlops(2*a->nz - m);
948: VecRestoreArray(xx,&x);
949: VecRestoreArray(yy,&y);
950: return(0);
951: }
955: PetscErrorCode MatMultAdd_SeqAIJ(Mat A,Vec xx,Vec yy,Vec zz)
956: {
957: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
958: PetscScalar *x,*y,*z,*aa;
960: PetscInt m = A->rmap.n,*aj,*ii;
961: #if !defined(PETSC_USE_FORTRAN_KERNEL_MULTADDAIJ)
962: PetscInt n,i,jrow,j,*ridx=PETSC_NULL;
963: PetscScalar sum;
964: PetscTruth usecprow=a->compressedrow.use;
965: #endif
968: VecGetArray(xx,&x);
969: VecGetArray(yy,&y);
970: if (zz != yy) {
971: VecGetArray(zz,&z);
972: } else {
973: z = y;
974: }
976: aj = a->j;
977: aa = a->a;
978: ii = a->i;
979: #if defined(PETSC_USE_FORTRAN_KERNEL_MULTADDAIJ)
980: fortranmultaddaij_(&m,x,ii,aj,aa,y,z);
981: #else
982: if (usecprow){ /* use compressed row format */
983: if (zz != yy){
984: PetscMemcpy(z,y,m*sizeof(PetscScalar));
985: }
986: m = a->compressedrow.nrows;
987: ii = a->compressedrow.i;
988: ridx = a->compressedrow.rindex;
989: for (i=0; i<m; i++){
990: n = ii[i+1] - ii[i];
991: aj = a->j + ii[i];
992: aa = a->a + ii[i];
993: sum = y[*ridx];
994: for (j=0; j<n; j++) sum += (*aa++)*x[*aj++];
995: z[*ridx++] = sum;
996: }
997: } else { /* do not use compressed row format */
998: for (i=0; i<m; i++) {
999: jrow = ii[i];
1000: n = ii[i+1] - jrow;
1001: sum = y[i];
1002: for (j=0; j<n; j++) {
1003: sum += aa[jrow]*x[aj[jrow]]; jrow++;
1004: }
1005: z[i] = sum;
1006: }
1007: }
1008: #endif
1009: PetscLogFlops(2*a->nz);
1010: VecRestoreArray(xx,&x);
1011: VecRestoreArray(yy,&y);
1012: if (zz != yy) {
1013: VecRestoreArray(zz,&z);
1014: }
1015: return(0);
1016: }
1018: /*
1019: Adds diagonal pointers to sparse matrix structure.
1020: */
1023: PetscErrorCode MatMarkDiagonal_SeqAIJ(Mat A)
1024: {
1025: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1027: PetscInt i,j,m = A->rmap.n;
1030: if (!a->diag) {
1031: PetscMalloc(m*sizeof(PetscInt),&a->diag);
1032: }
1033: for (i=0; i<A->rmap.n; i++) {
1034: a->diag[i] = a->i[i+1];
1035: for (j=a->i[i]; j<a->i[i+1]; j++) {
1036: if (a->j[j] == i) {
1037: a->diag[i] = j;
1038: break;
1039: }
1040: }
1041: }
1042: return(0);
1043: }
1045: /*
1046: Checks for missing diagonals
1047: */
1050: PetscErrorCode MatMissingDiagonal_SeqAIJ(Mat A,PetscTruth *missing,PetscInt *d)
1051: {
1052: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1053: PetscInt *diag,*jj = a->j,i;
1056: *missing = PETSC_FALSE;
1057: if (A->rmap.n > 0 && !jj) {
1058: *missing = PETSC_TRUE;
1059: if (d) *d = 0;
1060: PetscInfo(A,"Matrix has no entries therefor is missing diagonal");
1061: } else {
1062: diag = a->diag;
1063: for (i=0; i<A->rmap.n; i++) {
1064: if (jj[diag[i]] != i) {
1065: *missing = PETSC_TRUE;
1066: if (d) *d = i;
1067: PetscInfo1(A,"Matrix is missing diagonal number %D",i);
1068: }
1069: }
1070: }
1071: return(0);
1072: }
1076: PetscErrorCode MatRelax_SeqAIJ(Mat A,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,PetscInt its,PetscInt lits,Vec xx)
1077: {
1078: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1079: PetscScalar *x,d,*xs,sum,*t,scale,*idiag=0,*mdiag;
1080: const PetscScalar *v = a->a, *b, *bs,*xb, *ts;
1081: PetscErrorCode ierr;
1082: PetscInt n = A->cmap.n,m = A->rmap.n,i;
1083: const PetscInt *idx,*diag;
1086: its = its*lits;
1087: if (its <= 0) SETERRQ2(PETSC_ERR_ARG_WRONG,"Relaxation requires global its %D and local its %D both positive",its,lits);
1089: diag = a->diag;
1090: if (!a->idiag) {
1091: PetscMalloc(3*m*sizeof(PetscScalar),&a->idiag);
1092: a->ssor = a->idiag + m;
1093: mdiag = a->ssor + m;
1094: v = a->a;
1096: /* this is wrong when fshift omega changes each iteration */
1097: if (omega == 1.0 && !fshift) {
1098: for (i=0; i<m; i++) {
1099: mdiag[i] = v[diag[i]];
1100: if (!PetscAbsScalar(mdiag[i])) SETERRQ1(PETSC_ERR_ARG_INCOMP,"Zero diagonal on row %D",i);
1101: a->idiag[i] = 1.0/v[diag[i]];
1102: }
1103: PetscLogFlops(m);
1104: } else {
1105: for (i=0; i<m; i++) {
1106: mdiag[i] = v[diag[i]];
1107: a->idiag[i] = omega/(fshift + v[diag[i]]);
1108: }
1109: PetscLogFlops(2*m);
1110: }
1111: }
1112: t = a->ssor;
1113: idiag = a->idiag;
1114: mdiag = a->idiag + 2*m;
1116: VecGetArray(xx,&x);
1117: if (xx != bb) {
1118: VecGetArray(bb,(PetscScalar**)&b);
1119: } else {
1120: b = x;
1121: }
1123: /* We count flops by assuming the upper triangular and lower triangular parts have the same number of nonzeros */
1124: xs = x;
1125: if (flag == SOR_APPLY_UPPER) {
1126: /* apply (U + D/omega) to the vector */
1127: bs = b;
1128: for (i=0; i<m; i++) {
1129: d = fshift + a->a[diag[i]];
1130: n = a->i[i+1] - diag[i] - 1;
1131: idx = a->j + diag[i] + 1;
1132: v = a->a + diag[i] + 1;
1133: sum = b[i]*d/omega;
1134: SPARSEDENSEDOT(sum,bs,v,idx,n);
1135: x[i] = sum;
1136: }
1137: VecRestoreArray(xx,&x);
1138: if (bb != xx) {VecRestoreArray(bb,(PetscScalar**)&b);}
1139: PetscLogFlops(a->nz);
1140: return(0);
1141: }
1144: /* Let A = L + U + D; where L is lower trianglar,
1145: U is upper triangular, E is diagonal; This routine applies
1147: (L + E)^{-1} A (U + E)^{-1}
1149: to a vector efficiently using Eisenstat's trick. This is for
1150: the case of SSOR preconditioner, so E is D/omega where omega
1151: is the relaxation factor.
1152: */
1154: if (flag == SOR_APPLY_LOWER) {
1155: SETERRQ(PETSC_ERR_SUP,"SOR_APPLY_LOWER is not implemented");
1156: } else if (flag & SOR_EISENSTAT) {
1157: /* Let A = L + U + D; where L is lower trianglar,
1158: U is upper triangular, E is diagonal; This routine applies
1160: (L + E)^{-1} A (U + E)^{-1}
1162: to a vector efficiently using Eisenstat's trick. This is for
1163: the case of SSOR preconditioner, so E is D/omega where omega
1164: is the relaxation factor.
1165: */
1166: scale = (2.0/omega) - 1.0;
1168: /* x = (E + U)^{-1} b */
1169: for (i=m-1; i>=0; i--) {
1170: n = a->i[i+1] - diag[i] - 1;
1171: idx = a->j + diag[i] + 1;
1172: v = a->a + diag[i] + 1;
1173: sum = b[i];
1174: SPARSEDENSEMDOT(sum,xs,v,idx,n);
1175: x[i] = sum*idiag[i];
1176: }
1178: /* t = b - (2*E - D)x */
1179: v = a->a;
1180: for (i=0; i<m; i++) { t[i] = b[i] - scale*(v[*diag++])*x[i]; }
1182: /* t = (E + L)^{-1}t */
1183: ts = t;
1184: diag = a->diag;
1185: for (i=0; i<m; i++) {
1186: n = diag[i] - a->i[i];
1187: idx = a->j + a->i[i];
1188: v = a->a + a->i[i];
1189: sum = t[i];
1190: SPARSEDENSEMDOT(sum,ts,v,idx,n);
1191: t[i] = sum*idiag[i];
1192: /* x = x + t */
1193: x[i] += t[i];
1194: }
1196: PetscLogFlops(6*m-1 + 2*a->nz);
1197: VecRestoreArray(xx,&x);
1198: if (bb != xx) {VecRestoreArray(bb,(PetscScalar**)&b);}
1199: return(0);
1200: }
1201: if (flag & SOR_ZERO_INITIAL_GUESS) {
1202: if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP){
1203: #if defined(PETSC_USE_FORTRAN_KERNEL_RELAXAIJ)
1204: fortranrelaxaijforwardzero_(&m,&omega,x,a->i,a->j,(PetscInt*)diag,idiag,a->a,(void*)b);
1205: #else
1206: for (i=0; i<m; i++) {
1207: n = diag[i] - a->i[i];
1208: idx = a->j + a->i[i];
1209: v = a->a + a->i[i];
1210: sum = b[i];
1211: SPARSEDENSEMDOT(sum,xs,v,idx,n);
1212: x[i] = sum*idiag[i];
1213: }
1214: #endif
1215: xb = x;
1216: PetscLogFlops(a->nz);
1217: } else xb = b;
1218: if ((flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) &&
1219: (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP)) {
1220: for (i=0; i<m; i++) {
1221: x[i] *= mdiag[i];
1222: }
1223: PetscLogFlops(m);
1224: }
1225: if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP){
1226: #if defined(PETSC_USE_FORTRAN_KERNEL_RELAXAIJ)
1227: fortranrelaxaijbackwardzero_(&m,&omega,x,a->i,a->j,(PetscInt*)diag,idiag,a->a,(void*)xb);
1228: #else
1229: for (i=m-1; i>=0; i--) {
1230: n = a->i[i+1] - diag[i] - 1;
1231: idx = a->j + diag[i] + 1;
1232: v = a->a + diag[i] + 1;
1233: sum = xb[i];
1234: SPARSEDENSEMDOT(sum,xs,v,idx,n);
1235: x[i] = sum*idiag[i];
1236: }
1237: #endif
1238: PetscLogFlops(a->nz);
1239: }
1240: its--;
1241: }
1242: while (its--) {
1243: if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP){
1244: #if defined(PETSC_USE_FORTRAN_KERNEL_RELAXAIJ)
1245: fortranrelaxaijforward_(&m,&omega,x,a->i,a->j,(PetscInt*)diag,a->a,(void*)b);
1246: #else
1247: for (i=0; i<m; i++) {
1248: n = a->i[i+1] - a->i[i];
1249: idx = a->j + a->i[i];
1250: v = a->a + a->i[i];
1251: sum = b[i];
1252: SPARSEDENSEMDOT(sum,xs,v,idx,n);
1253: x[i] = (1. - omega)*x[i] + (sum + mdiag[i]*x[i])*idiag[i];
1254: }
1255: #endif
1256: PetscLogFlops(a->nz);
1257: }
1258: if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP){
1259: #if defined(PETSC_USE_FORTRAN_KERNEL_RELAXAIJ)
1260: fortranrelaxaijbackward_(&m,&omega,x,a->i,a->j,(PetscInt*)diag,a->a,(void*)b);
1261: #else
1262: for (i=m-1; i>=0; i--) {
1263: n = a->i[i+1] - a->i[i];
1264: idx = a->j + a->i[i];
1265: v = a->a + a->i[i];
1266: sum = b[i];
1267: SPARSEDENSEMDOT(sum,xs,v,idx,n);
1268: x[i] = (1. - omega)*x[i] + (sum + mdiag[i]*x[i])*idiag[i];
1269: }
1270: #endif
1271: PetscLogFlops(a->nz);
1272: }
1273: }
1274: VecRestoreArray(xx,&x);
1275: if (bb != xx) {VecRestoreArray(bb,(PetscScalar**)&b);}
1276: return(0);
1277: }
1281: PetscErrorCode MatGetInfo_SeqAIJ(Mat A,MatInfoType flag,MatInfo *info)
1282: {
1283: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1286: info->rows_global = (double)A->rmap.n;
1287: info->columns_global = (double)A->cmap.n;
1288: info->rows_local = (double)A->rmap.n;
1289: info->columns_local = (double)A->cmap.n;
1290: info->block_size = 1.0;
1291: info->nz_allocated = (double)a->maxnz;
1292: info->nz_used = (double)a->nz;
1293: info->nz_unneeded = (double)(a->maxnz - a->nz);
1294: info->assemblies = (double)A->num_ass;
1295: info->mallocs = (double)a->reallocs;
1296: info->memory = A->mem;
1297: if (A->factor) {
1298: info->fill_ratio_given = A->info.fill_ratio_given;
1299: info->fill_ratio_needed = A->info.fill_ratio_needed;
1300: info->factor_mallocs = A->info.factor_mallocs;
1301: } else {
1302: info->fill_ratio_given = 0;
1303: info->fill_ratio_needed = 0;
1304: info->factor_mallocs = 0;
1305: }
1306: return(0);
1307: }
1311: PetscErrorCode MatZeroRows_SeqAIJ(Mat A,PetscInt N,const PetscInt rows[],PetscScalar diag)
1312: {
1313: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1314: PetscInt i,m = A->rmap.n - 1,d;
1316: PetscTruth missing;
1319: if (a->keepzeroedrows) {
1320: for (i=0; i<N; i++) {
1321: if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]);
1322: PetscMemzero(&a->a[a->i[rows[i]]],a->ilen[rows[i]]*sizeof(PetscScalar));
1323: }
1324: if (diag != 0.0) {
1325: MatMissingDiagonal_SeqAIJ(A,&missing,&d);
1326: if (missing) SETERRQ1(PETSC_ERR_ARG_WRONGSTATE,"Matrix is missing diagonal entry in row %D",d);
1327: for (i=0; i<N; i++) {
1328: a->a[a->diag[rows[i]]] = diag;
1329: }
1330: }
1331: A->same_nonzero = PETSC_TRUE;
1332: } else {
1333: if (diag != 0.0) {
1334: for (i=0; i<N; i++) {
1335: if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]);
1336: if (a->ilen[rows[i]] > 0) {
1337: a->ilen[rows[i]] = 1;
1338: a->a[a->i[rows[i]]] = diag;
1339: a->j[a->i[rows[i]]] = rows[i];
1340: } else { /* in case row was completely empty */
1341: MatSetValues_SeqAIJ(A,1,&rows[i],1,&rows[i],&diag,INSERT_VALUES);
1342: }
1343: }
1344: } else {
1345: for (i=0; i<N; i++) {
1346: if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range", rows[i]);
1347: a->ilen[rows[i]] = 0;
1348: }
1349: }
1350: A->same_nonzero = PETSC_FALSE;
1351: }
1352: MatAssemblyEnd_SeqAIJ(A,MAT_FINAL_ASSEMBLY);
1353: return(0);
1354: }
1358: PetscErrorCode MatGetRow_SeqAIJ(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v)
1359: {
1360: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1361: PetscInt *itmp;
1364: if (row < 0 || row >= A->rmap.n) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"Row %D out of range",row);
1366: *nz = a->i[row+1] - a->i[row];
1367: if (v) *v = a->a + a->i[row];
1368: if (idx) {
1369: itmp = a->j + a->i[row];
1370: if (*nz) {
1371: *idx = itmp;
1372: }
1373: else *idx = 0;
1374: }
1375: return(0);
1376: }
1378: /* remove this function? */
1381: PetscErrorCode MatRestoreRow_SeqAIJ(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v)
1382: {
1384: return(0);
1385: }
1389: PetscErrorCode MatNorm_SeqAIJ(Mat A,NormType type,PetscReal *nrm)
1390: {
1391: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1392: PetscScalar *v = a->a;
1393: PetscReal sum = 0.0;
1395: PetscInt i,j;
1398: if (type == NORM_FROBENIUS) {
1399: for (i=0; i<a->nz; i++) {
1400: #if defined(PETSC_USE_COMPLEX)
1401: sum += PetscRealPart(PetscConj(*v)*(*v)); v++;
1402: #else
1403: sum += (*v)*(*v); v++;
1404: #endif
1405: }
1406: *nrm = sqrt(sum);
1407: } else if (type == NORM_1) {
1408: PetscReal *tmp;
1409: PetscInt *jj = a->j;
1410: PetscMalloc((A->cmap.n+1)*sizeof(PetscReal),&tmp);
1411: PetscMemzero(tmp,A->cmap.n*sizeof(PetscReal));
1412: *nrm = 0.0;
1413: for (j=0; j<a->nz; j++) {
1414: tmp[*jj++] += PetscAbsScalar(*v); v++;
1415: }
1416: for (j=0; j<A->cmap.n; j++) {
1417: if (tmp[j] > *nrm) *nrm = tmp[j];
1418: }
1419: PetscFree(tmp);
1420: } else if (type == NORM_INFINITY) {
1421: *nrm = 0.0;
1422: for (j=0; j<A->rmap.n; j++) {
1423: v = a->a + a->i[j];
1424: sum = 0.0;
1425: for (i=0; i<a->i[j+1]-a->i[j]; i++) {
1426: sum += PetscAbsScalar(*v); v++;
1427: }
1428: if (sum > *nrm) *nrm = sum;
1429: }
1430: } else {
1431: SETERRQ(PETSC_ERR_SUP,"No support for two norm");
1432: }
1433: return(0);
1434: }
1438: PetscErrorCode MatTranspose_SeqAIJ(Mat A,Mat *B)
1439: {
1440: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1441: Mat C;
1443: PetscInt i,*aj = a->j,*ai = a->i,m = A->rmap.n,len,*col;
1444: PetscScalar *array = a->a;
1447: if (!B && m != A->cmap.n) SETERRQ(PETSC_ERR_ARG_SIZ,"Square matrix only for in-place");
1448: PetscMalloc((1+A->cmap.n)*sizeof(PetscInt),&col);
1449: PetscMemzero(col,(1+A->cmap.n)*sizeof(PetscInt));
1450:
1451: for (i=0; i<ai[m]; i++) col[aj[i]] += 1;
1452: MatCreate(A->comm,&C);
1453: MatSetSizes(C,A->cmap.n,m,A->cmap.n,m);
1454: MatSetType(C,A->type_name);
1455: MatSeqAIJSetPreallocation_SeqAIJ(C,0,col);
1456: PetscFree(col);
1457: for (i=0; i<m; i++) {
1458: len = ai[i+1]-ai[i];
1459: MatSetValues_SeqAIJ(C,len,aj,1,&i,array,INSERT_VALUES);
1460: array += len;
1461: aj += len;
1462: }
1464: MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);
1465: MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);
1467: if (B) {
1468: *B = C;
1469: } else {
1470: MatHeaderCopy(A,C);
1471: }
1472: return(0);
1473: }
1478: PetscErrorCode MatIsTranspose_SeqAIJ(Mat A,Mat B,PetscReal tol,PetscTruth *f)
1479: {
1480: Mat_SeqAIJ *aij = (Mat_SeqAIJ *) A->data,*bij = (Mat_SeqAIJ*) A->data;
1481: PetscInt *adx,*bdx,*aii,*bii,*aptr,*bptr; PetscScalar *va,*vb;
1483: PetscInt ma,na,mb,nb, i;
1486: bij = (Mat_SeqAIJ *) B->data;
1487:
1488: MatGetSize(A,&ma,&na);
1489: MatGetSize(B,&mb,&nb);
1490: if (ma!=nb || na!=mb){
1491: *f = PETSC_FALSE;
1492: return(0);
1493: }
1494: aii = aij->i; bii = bij->i;
1495: adx = aij->j; bdx = bij->j;
1496: va = aij->a; vb = bij->a;
1497: PetscMalloc(ma*sizeof(PetscInt),&aptr);
1498: PetscMalloc(mb*sizeof(PetscInt),&bptr);
1499: for (i=0; i<ma; i++) aptr[i] = aii[i];
1500: for (i=0; i<mb; i++) bptr[i] = bii[i];
1502: *f = PETSC_TRUE;
1503: for (i=0; i<ma; i++) {
1504: while (aptr[i]<aii[i+1]) {
1505: PetscInt idc,idr;
1506: PetscScalar vc,vr;
1507: /* column/row index/value */
1508: idc = adx[aptr[i]];
1509: idr = bdx[bptr[idc]];
1510: vc = va[aptr[i]];
1511: vr = vb[bptr[idc]];
1512: if (i!=idr || PetscAbsScalar(vc-vr) > tol) {
1513: *f = PETSC_FALSE;
1514: goto done;
1515: } else {
1516: aptr[i]++;
1517: if (B || i!=idc) bptr[idc]++;
1518: }
1519: }
1520: }
1521: done:
1522: PetscFree(aptr);
1523: if (B) {
1524: PetscFree(bptr);
1525: }
1526: return(0);
1527: }
1532: PetscErrorCode MatIsSymmetric_SeqAIJ(Mat A,PetscReal tol,PetscTruth *f)
1533: {
1536: MatIsTranspose_SeqAIJ(A,A,tol,f);
1537: return(0);
1538: }
1542: PetscErrorCode MatDiagonalScale_SeqAIJ(Mat A,Vec ll,Vec rr)
1543: {
1544: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1545: PetscScalar *l,*r,x,*v;
1547: PetscInt i,j,m = A->rmap.n,n = A->cmap.n,M,nz = a->nz,*jj;
1550: if (ll) {
1551: /* The local size is used so that VecMPI can be passed to this routine
1552: by MatDiagonalScale_MPIAIJ */
1553: VecGetLocalSize(ll,&m);
1554: if (m != A->rmap.n) SETERRQ(PETSC_ERR_ARG_SIZ,"Left scaling vector wrong length");
1555: VecGetArray(ll,&l);
1556: v = a->a;
1557: for (i=0; i<m; i++) {
1558: x = l[i];
1559: M = a->i[i+1] - a->i[i];
1560: for (j=0; j<M; j++) { (*v++) *= x;}
1561: }
1562: VecRestoreArray(ll,&l);
1563: PetscLogFlops(nz);
1564: }
1565: if (rr) {
1566: VecGetLocalSize(rr,&n);
1567: if (n != A->cmap.n) SETERRQ(PETSC_ERR_ARG_SIZ,"Right scaling vector wrong length");
1568: VecGetArray(rr,&r);
1569: v = a->a; jj = a->j;
1570: for (i=0; i<nz; i++) {
1571: (*v++) *= r[*jj++];
1572: }
1573: VecRestoreArray(rr,&r);
1574: PetscLogFlops(nz);
1575: }
1576: return(0);
1577: }
1581: PetscErrorCode MatGetSubMatrix_SeqAIJ(Mat A,IS isrow,IS iscol,PetscInt csize,MatReuse scall,Mat *B)
1582: {
1583: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*c;
1585: PetscInt *smap,i,k,kstart,kend,oldcols = A->cmap.n,*lens;
1586: PetscInt row,mat_i,*mat_j,tcol,first,step,*mat_ilen,sum,lensi;
1587: PetscInt *irow,*icol,nrows,ncols;
1588: PetscInt *starts,*j_new,*i_new,*aj = a->j,*ai = a->i,ii,*ailen = a->ilen;
1589: PetscScalar *a_new,*mat_a;
1590: Mat C;
1591: PetscTruth stride;
1594: ISSorted(isrow,(PetscTruth*)&i);
1595: if (!i) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"ISrow is not sorted");
1596: ISSorted(iscol,(PetscTruth*)&i);
1597: if (!i) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"IScol is not sorted");
1599: ISGetIndices(isrow,&irow);
1600: ISGetLocalSize(isrow,&nrows);
1601: ISGetLocalSize(iscol,&ncols);
1603: ISStrideGetInfo(iscol,&first,&step);
1604: ISStride(iscol,&stride);
1605: if (stride && step == 1) {
1606: /* special case of contiguous rows */
1607: PetscMalloc((2*nrows+1)*sizeof(PetscInt),&lens);
1608: starts = lens + nrows;
1609: /* loop over new rows determining lens and starting points */
1610: for (i=0; i<nrows; i++) {
1611: kstart = ai[irow[i]];
1612: kend = kstart + ailen[irow[i]];
1613: for (k=kstart; k<kend; k++) {
1614: if (aj[k] >= first) {
1615: starts[i] = k;
1616: break;
1617: }
1618: }
1619: sum = 0;
1620: while (k < kend) {
1621: if (aj[k++] >= first+ncols) break;
1622: sum++;
1623: }
1624: lens[i] = sum;
1625: }
1626: /* create submatrix */
1627: if (scall == MAT_REUSE_MATRIX) {
1628: PetscInt n_cols,n_rows;
1629: MatGetSize(*B,&n_rows,&n_cols);
1630: if (n_rows != nrows || n_cols != ncols) SETERRQ(PETSC_ERR_ARG_SIZ,"Reused submatrix wrong size");
1631: MatZeroEntries(*B);
1632: C = *B;
1633: } else {
1634: MatCreate(A->comm,&C);
1635: MatSetSizes(C,nrows,ncols,PETSC_DETERMINE,PETSC_DETERMINE);
1636: MatSetType(C,A->type_name);
1637: MatSeqAIJSetPreallocation_SeqAIJ(C,0,lens);
1638: }
1639: c = (Mat_SeqAIJ*)C->data;
1641: /* loop over rows inserting into submatrix */
1642: a_new = c->a;
1643: j_new = c->j;
1644: i_new = c->i;
1646: for (i=0; i<nrows; i++) {
1647: ii = starts[i];
1648: lensi = lens[i];
1649: for (k=0; k<lensi; k++) {
1650: *j_new++ = aj[ii+k] - first;
1651: }
1652: PetscMemcpy(a_new,a->a + starts[i],lensi*sizeof(PetscScalar));
1653: a_new += lensi;
1654: i_new[i+1] = i_new[i] + lensi;
1655: c->ilen[i] = lensi;
1656: }
1657: PetscFree(lens);
1658: } else {
1659: ISGetIndices(iscol,&icol);
1660: PetscMalloc((1+oldcols)*sizeof(PetscInt),&smap);
1661:
1662: PetscMalloc((1+nrows)*sizeof(PetscInt),&lens);
1663: PetscMemzero(smap,oldcols*sizeof(PetscInt));
1664: for (i=0; i<ncols; i++) smap[icol[i]] = i+1;
1665: /* determine lens of each row */
1666: for (i=0; i<nrows; i++) {
1667: kstart = ai[irow[i]];
1668: kend = kstart + a->ilen[irow[i]];
1669: lens[i] = 0;
1670: for (k=kstart; k<kend; k++) {
1671: if (smap[aj[k]]) {
1672: lens[i]++;
1673: }
1674: }
1675: }
1676: /* Create and fill new matrix */
1677: if (scall == MAT_REUSE_MATRIX) {
1678: PetscTruth equal;
1680: c = (Mat_SeqAIJ *)((*B)->data);
1681: if ((*B)->rmap.n != nrows || (*B)->cmap.n != ncols) SETERRQ(PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. wrong size");
1682: PetscMemcmp(c->ilen,lens,(*B)->rmap.n*sizeof(PetscInt),&equal);
1683: if (!equal) {
1684: SETERRQ(PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. wrong no of nonzeros");
1685: }
1686: PetscMemzero(c->ilen,(*B)->rmap.n*sizeof(PetscInt));
1687: C = *B;
1688: } else {
1689: MatCreate(A->comm,&C);
1690: MatSetSizes(C,nrows,ncols,PETSC_DETERMINE,PETSC_DETERMINE);
1691: MatSetType(C,A->type_name);
1692: MatSeqAIJSetPreallocation_SeqAIJ(C,0,lens);
1693: }
1694: c = (Mat_SeqAIJ *)(C->data);
1695: for (i=0; i<nrows; i++) {
1696: row = irow[i];
1697: kstart = ai[row];
1698: kend = kstart + a->ilen[row];
1699: mat_i = c->i[i];
1700: mat_j = c->j + mat_i;
1701: mat_a = c->a + mat_i;
1702: mat_ilen = c->ilen + i;
1703: for (k=kstart; k<kend; k++) {
1704: if ((tcol=smap[a->j[k]])) {
1705: *mat_j++ = tcol - 1;
1706: *mat_a++ = a->a[k];
1707: (*mat_ilen)++;
1709: }
1710: }
1711: }
1712: /* Free work space */
1713: ISRestoreIndices(iscol,&icol);
1714: PetscFree(smap);
1715: PetscFree(lens);
1716: }
1717: MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);
1718: MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);
1720: ISRestoreIndices(isrow,&irow);
1721: *B = C;
1722: return(0);
1723: }
1725: /*
1726: */
1729: PetscErrorCode MatILUFactor_SeqAIJ(Mat inA,IS row,IS col,MatFactorInfo *info)
1730: {
1731: Mat_SeqAIJ *a = (Mat_SeqAIJ*)inA->data;
1733: Mat outA;
1734: PetscTruth row_identity,col_identity;
1737: if (info->levels != 0) SETERRQ(PETSC_ERR_SUP,"Only levels=0 supported for in-place ilu");
1738: ISIdentity(row,&row_identity);
1739: ISIdentity(col,&col_identity);
1741: outA = inA;
1742: inA->factor = FACTOR_LU;
1743: a->row = row;
1744: a->col = col;
1745: PetscObjectReference((PetscObject)row);
1746: PetscObjectReference((PetscObject)col);
1748: /* Create the inverse permutation so that it can be used in MatLUFactorNumeric() */
1749: if (a->icol) {ISDestroy(a->icol);} /* need to remove old one */
1750: ISInvertPermutation(col,PETSC_DECIDE,&a->icol);
1751: PetscLogObjectParent(inA,a->icol);
1753: if (!a->solve_work) { /* this matrix may have been factored before */
1754: PetscMalloc((inA->rmap.n+1)*sizeof(PetscScalar),&a->solve_work);
1755: }
1757: MatMarkDiagonal_SeqAIJ(inA);
1758: if (row_identity && col_identity) {
1759: MatLUFactorNumeric_SeqAIJ(inA,info,&outA);
1760: } else {
1761: MatLUFactorNumeric_SeqAIJ_InplaceWithPerm(inA,info,&outA);
1762: }
1763: return(0);
1764: }
1766: #include petscblaslapack.h
1769: PetscErrorCode MatScale_SeqAIJ(Mat inA,PetscScalar alpha)
1770: {
1771: Mat_SeqAIJ *a = (Mat_SeqAIJ*)inA->data;
1772: PetscBLASInt bnz = (PetscBLASInt)a->nz,one = 1;
1773: PetscScalar oalpha = alpha;
1778: BLASscal_(&bnz,&oalpha,a->a,&one);
1779: PetscLogFlops(a->nz);
1780: return(0);
1781: }
1785: PetscErrorCode MatGetSubMatrices_SeqAIJ(Mat A,PetscInt n,const IS irow[],const IS icol[],MatReuse scall,Mat *B[])
1786: {
1788: PetscInt i;
1791: if (scall == MAT_INITIAL_MATRIX) {
1792: PetscMalloc((n+1)*sizeof(Mat),B);
1793: }
1795: for (i=0; i<n; i++) {
1796: MatGetSubMatrix_SeqAIJ(A,irow[i],icol[i],PETSC_DECIDE,scall,&(*B)[i]);
1797: }
1798: return(0);
1799: }
1803: PetscErrorCode MatIncreaseOverlap_SeqAIJ(Mat A,PetscInt is_max,IS is[],PetscInt ov)
1804: {
1805: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1807: PetscInt row,i,j,k,l,m,n,*idx,*nidx,isz,val;
1808: PetscInt start,end,*ai,*aj;
1809: PetscBT table;
1812: m = A->rmap.n;
1813: ai = a->i;
1814: aj = a->j;
1816: if (ov < 0) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"illegal negative overlap value used");
1818: PetscMalloc((m+1)*sizeof(PetscInt),&nidx);
1819: PetscBTCreate(m,table);
1821: for (i=0; i<is_max; i++) {
1822: /* Initialize the two local arrays */
1823: isz = 0;
1824: PetscBTMemzero(m,table);
1825:
1826: /* Extract the indices, assume there can be duplicate entries */
1827: ISGetIndices(is[i],&idx);
1828: ISGetLocalSize(is[i],&n);
1829:
1830: /* Enter these into the temp arrays. I.e., mark table[row], enter row into new index */
1831: for (j=0; j<n ; ++j){
1832: if(!PetscBTLookupSet(table,idx[j])) { nidx[isz++] = idx[j];}
1833: }
1834: ISRestoreIndices(is[i],&idx);
1835: ISDestroy(is[i]);
1836:
1837: k = 0;
1838: for (j=0; j<ov; j++){ /* for each overlap */
1839: n = isz;
1840: for (; k<n ; k++){ /* do only those rows in nidx[k], which are not done yet */
1841: row = nidx[k];
1842: start = ai[row];
1843: end = ai[row+1];
1844: for (l = start; l<end ; l++){
1845: val = aj[l] ;
1846: if (!PetscBTLookupSet(table,val)) {nidx[isz++] = val;}
1847: }
1848: }
1849: }
1850: ISCreateGeneral(PETSC_COMM_SELF,isz,nidx,(is+i));
1851: }
1852: PetscBTDestroy(table);
1853: PetscFree(nidx);
1854: return(0);
1855: }
1857: /* -------------------------------------------------------------- */
1860: PetscErrorCode MatPermute_SeqAIJ(Mat A,IS rowp,IS colp,Mat *B)
1861: {
1862: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1864: PetscInt i,nz,m = A->rmap.n,n = A->cmap.n,*col;
1865: PetscInt *row,*cnew,j,*lens;
1866: IS icolp,irowp;
1867: PetscInt *cwork;
1868: PetscScalar *vwork;
1871: ISInvertPermutation(rowp,PETSC_DECIDE,&irowp);
1872: ISGetIndices(irowp,&row);
1873: ISInvertPermutation(colp,PETSC_DECIDE,&icolp);
1874: ISGetIndices(icolp,&col);
1875:
1876: /* determine lengths of permuted rows */
1877: PetscMalloc((m+1)*sizeof(PetscInt),&lens);
1878: for (i=0; i<m; i++) {
1879: lens[row[i]] = a->i[i+1] - a->i[i];
1880: }
1881: MatCreate(A->comm,B);
1882: MatSetSizes(*B,m,n,m,n);
1883: MatSetType(*B,A->type_name);
1884: MatSeqAIJSetPreallocation_SeqAIJ(*B,0,lens);
1885: PetscFree(lens);
1887: PetscMalloc(n*sizeof(PetscInt),&cnew);
1888: for (i=0; i<m; i++) {
1889: MatGetRow_SeqAIJ(A,i,&nz,&cwork,&vwork);
1890: for (j=0; j<nz; j++) { cnew[j] = col[cwork[j]];}
1891: MatSetValues_SeqAIJ(*B,1,&row[i],nz,cnew,vwork,INSERT_VALUES);
1892: MatRestoreRow_SeqAIJ(A,i,&nz,&cwork,&vwork);
1893: }
1894: PetscFree(cnew);
1895: (*B)->assembled = PETSC_FALSE;
1896: MatAssemblyBegin(*B,MAT_FINAL_ASSEMBLY);
1897: MatAssemblyEnd(*B,MAT_FINAL_ASSEMBLY);
1898: ISRestoreIndices(irowp,&row);
1899: ISRestoreIndices(icolp,&col);
1900: ISDestroy(irowp);
1901: ISDestroy(icolp);
1902: return(0);
1903: }
1907: PetscErrorCode MatCopy_SeqAIJ(Mat A,Mat B,MatStructure str)
1908: {
1912: /* If the two matrices have the same copy implementation, use fast copy. */
1913: if (str == SAME_NONZERO_PATTERN && (A->ops->copy == B->ops->copy)) {
1914: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1915: Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data;
1917: if (a->i[A->rmap.n] != b->i[B->rmap.n]) {
1918: SETERRQ(PETSC_ERR_ARG_INCOMP,"Number of nonzeros in two matrices are different");
1919: }
1920: PetscMemcpy(b->a,a->a,(a->i[A->rmap.n])*sizeof(PetscScalar));
1921: } else {
1922: MatCopy_Basic(A,B,str);
1923: }
1924: return(0);
1925: }
1929: PetscErrorCode MatSetUpPreallocation_SeqAIJ(Mat A)
1930: {
1934: MatSeqAIJSetPreallocation_SeqAIJ(A,PETSC_DEFAULT,0);
1935: return(0);
1936: }
1940: PetscErrorCode MatGetArray_SeqAIJ(Mat A,PetscScalar *array[])
1941: {
1942: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1944: *array = a->a;
1945: return(0);
1946: }
1950: PetscErrorCode MatRestoreArray_SeqAIJ(Mat A,PetscScalar *array[])
1951: {
1953: return(0);
1954: }
1958: PetscErrorCode MatFDColoringApply_SeqAIJ(Mat J,MatFDColoring coloring,Vec x1,MatStructure *flag,void *sctx)
1959: {
1960: PetscErrorCode (*f)(void*,Vec,Vec,void*) = (PetscErrorCode (*)(void*,Vec,Vec,void *))coloring->f;
1962: PetscInt k,N,start,end,l,row,col,srow,**vscaleforrow,m1,m2;
1963: PetscScalar dx,*y,*xx,*w3_array;
1964: PetscScalar *vscale_array;
1965: PetscReal epsilon = coloring->error_rel,umin = coloring->umin;
1966: Vec w1,w2,w3;
1967: void *fctx = coloring->fctx;
1968: PetscTruth flg;
1971: if (!coloring->w1) {
1972: VecDuplicate(x1,&coloring->w1);
1973: PetscLogObjectParent(coloring,coloring->w1);
1974: VecDuplicate(x1,&coloring->w2);
1975: PetscLogObjectParent(coloring,coloring->w2);
1976: VecDuplicate(x1,&coloring->w3);
1977: PetscLogObjectParent(coloring,coloring->w3);
1978: }
1979: w1 = coloring->w1; w2 = coloring->w2; w3 = coloring->w3;
1981: MatSetUnfactored(J);
1982: PetscOptionsHasName(coloring->prefix,"-mat_fd_coloring_dont_rezero",&flg);
1983: if (flg) {
1984: PetscInfo(coloring,"Not calling MatZeroEntries()\n");
1985: } else {
1986: PetscTruth assembled;
1987: MatAssembled(J,&assembled);
1988: if (assembled) {
1989: MatZeroEntries(J);
1990: }
1991: }
1993: VecGetOwnershipRange(x1,&start,&end);
1994: VecGetSize(x1,&N);
1996: /*
1997: This is a horrible, horrible, hack. See DMMGComputeJacobian_Multigrid() it inproperly sets
1998: coloring->F for the coarser grids from the finest
1999: */
2000: if (coloring->F) {
2001: VecGetLocalSize(coloring->F,&m1);
2002: VecGetLocalSize(w1,&m2);
2003: if (m1 != m2) {
2004: coloring->F = 0;
2005: }
2006: }
2008: if (coloring->F) {
2009: w1 = coloring->F;
2010: coloring->F = 0;
2011: } else {
2013: (*f)(sctx,x1,w1,fctx);
2015: }
2017: /*
2018: Compute all the scale factors and share with other processors
2019: */
2020: VecGetArray(x1,&xx);xx = xx - start;
2021: VecGetArray(coloring->vscale,&vscale_array);vscale_array = vscale_array - start;
2022: for (k=0; k<coloring->ncolors; k++) {
2023: /*
2024: Loop over each column associated with color adding the
2025: perturbation to the vector w3.
2026: */
2027: for (l=0; l<coloring->ncolumns[k]; l++) {
2028: col = coloring->columns[k][l]; /* column of the matrix we are probing for */
2029: dx = xx[col];
2030: if (dx == 0.0) dx = 1.0;
2031: #if !defined(PETSC_USE_COMPLEX)
2032: if (dx < umin && dx >= 0.0) dx = umin;
2033: else if (dx < 0.0 && dx > -umin) dx = -umin;
2034: #else
2035: if (PetscAbsScalar(dx) < umin && PetscRealPart(dx) >= 0.0) dx = umin;
2036: else if (PetscRealPart(dx) < 0.0 && PetscAbsScalar(dx) < umin) dx = -umin;
2037: #endif
2038: dx *= epsilon;
2039: vscale_array[col] = 1.0/dx;
2040: }
2041: }
2042: vscale_array = vscale_array + start;VecRestoreArray(coloring->vscale,&vscale_array);
2043: VecGhostUpdateBegin(coloring->vscale,INSERT_VALUES,SCATTER_FORWARD);
2044: VecGhostUpdateEnd(coloring->vscale,INSERT_VALUES,SCATTER_FORWARD);
2046: /* VecView(coloring->vscale,PETSC_VIEWER_STDOUT_WORLD);
2047: VecView(x1,PETSC_VIEWER_STDOUT_WORLD);*/
2049: if (coloring->vscaleforrow) vscaleforrow = coloring->vscaleforrow;
2050: else vscaleforrow = coloring->columnsforrow;
2052: VecGetArray(coloring->vscale,&vscale_array);
2053: /*
2054: Loop over each color
2055: */
2056: for (k=0; k<coloring->ncolors; k++) {
2057: coloring->currentcolor = k;
2058: VecCopy(x1,w3);
2059: VecGetArray(w3,&w3_array);w3_array = w3_array - start;
2060: /*
2061: Loop over each column associated with color adding the
2062: perturbation to the vector w3.
2063: */
2064: for (l=0; l<coloring->ncolumns[k]; l++) {
2065: col = coloring->columns[k][l]; /* column of the matrix we are probing for */
2066: dx = xx[col];
2067: if (dx == 0.0) dx = 1.0;
2068: #if !defined(PETSC_USE_COMPLEX)
2069: if (dx < umin && dx >= 0.0) dx = umin;
2070: else if (dx < 0.0 && dx > -umin) dx = -umin;
2071: #else
2072: if (PetscAbsScalar(dx) < umin && PetscRealPart(dx) >= 0.0) dx = umin;
2073: else if (PetscRealPart(dx) < 0.0 && PetscAbsScalar(dx) < umin) dx = -umin;
2074: #endif
2075: dx *= epsilon;
2076: if (!PetscAbsScalar(dx)) SETERRQ(PETSC_ERR_PLIB,"Computed 0 differencing parameter");
2077: w3_array[col] += dx;
2078: }
2079: w3_array = w3_array + start; VecRestoreArray(w3,&w3_array);
2081: /*
2082: Evaluate function at x1 + dx (here dx is a vector of perturbations)
2083: */
2086: (*f)(sctx,w3,w2,fctx);
2088: VecAXPY(w2,-1.0,w1);
2090: /*
2091: Loop over rows of vector, putting results into Jacobian matrix
2092: */
2093: VecGetArray(w2,&y);
2094: for (l=0; l<coloring->nrows[k]; l++) {
2095: row = coloring->rows[k][l];
2096: col = coloring->columnsforrow[k][l];
2097: y[row] *= vscale_array[vscaleforrow[k][l]];
2098: srow = row + start;
2099: MatSetValues_SeqAIJ(J,1,&srow,1,&col,y+row,INSERT_VALUES);
2100: }
2101: VecRestoreArray(w2,&y);
2102: }
2103: coloring->currentcolor = k;
2104: VecRestoreArray(coloring->vscale,&vscale_array);
2105: xx = xx + start; VecRestoreArray(x1,&xx);
2106: MatAssemblyBegin(J,MAT_FINAL_ASSEMBLY);
2107: MatAssemblyEnd(J,MAT_FINAL_ASSEMBLY);
2108: return(0);
2109: }
2111: #include petscblaslapack.h
2114: PetscErrorCode MatAXPY_SeqAIJ(Mat Y,PetscScalar a,Mat X,MatStructure str)
2115: {
2117: PetscInt i;
2118: Mat_SeqAIJ *x = (Mat_SeqAIJ *)X->data,*y = (Mat_SeqAIJ *)Y->data;
2119: PetscBLASInt one=1,bnz = (PetscBLASInt)x->nz;
2122: if (str == SAME_NONZERO_PATTERN) {
2123: PetscScalar alpha = a;
2124: BLASaxpy_(&bnz,&alpha,x->a,&one,y->a,&one);
2125: } else if (str == SUBSET_NONZERO_PATTERN) { /* nonzeros of X is a subset of Y's */
2126: if (y->xtoy && y->XtoY != X) {
2127: PetscFree(y->xtoy);
2128: MatDestroy(y->XtoY);
2129: }
2130: if (!y->xtoy) { /* get xtoy */
2131: MatAXPYGetxtoy_Private(X->rmap.n,x->i,x->j,PETSC_NULL, y->i,y->j,PETSC_NULL, &y->xtoy);
2132: y->XtoY = X;
2133: }
2134: for (i=0; i<x->nz; i++) y->a[y->xtoy[i]] += a*(x->a[i]);
2135: PetscInfo3(0,"ratio of nnz(X)/nnz(Y): %d/%d = %G\n",x->nz,y->nz,(PetscReal)(x->nz)/y->nz);
2136: } else {
2137: MatAXPY_Basic(Y,a,X,str);
2138: }
2139: return(0);
2140: }
2144: PetscErrorCode MatSetBlockSize_SeqAIJ(Mat A,PetscInt bs)
2145: {
2147: return(0);
2148: }
2152: PetscErrorCode MatConjugate_SeqAIJ(Mat mat)
2153: {
2154: #if defined(PETSC_USE_COMPLEX)
2155: Mat_SeqAIJ *aij = (Mat_SeqAIJ *)mat->data;
2156: PetscInt i,nz;
2157: PetscScalar *a;
2160: nz = aij->nz;
2161: a = aij->a;
2162: for (i=0; i<nz; i++) {
2163: a[i] = PetscConj(a[i]);
2164: }
2165: #else
2167: #endif
2168: return(0);
2169: }
2173: PetscErrorCode MatGetRowMax_SeqAIJ(Mat A,Vec v)
2174: {
2175: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2177: PetscInt i,j,m = A->rmap.n,*ai,*aj,ncols,n;
2178: PetscReal atmp;
2179: PetscScalar *x,zero = 0.0;
2180: MatScalar *aa;
2183: if (A->factor) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
2184: aa = a->a;
2185: ai = a->i;
2186: aj = a->j;
2188: VecSet(v,zero);
2189: VecGetArray(v,&x);
2190: VecGetLocalSize(v,&n);
2191: if (n != A->rmap.n) SETERRQ(PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector");
2192: for (i=0; i<m; i++) {
2193: ncols = ai[1] - ai[0]; ai++;
2194: for (j=0; j<ncols; j++){
2195: atmp = PetscAbsScalar(*aa); aa++;
2196: if (PetscAbsScalar(x[i]) < atmp) x[i] = atmp;
2197: aj++;
2198: }
2199: }
2200: VecRestoreArray(v,&x);
2201: return(0);
2202: }
2204: /* -------------------------------------------------------------------*/
2205: static struct _MatOps MatOps_Values = {MatSetValues_SeqAIJ,
2206: MatGetRow_SeqAIJ,
2207: MatRestoreRow_SeqAIJ,
2208: MatMult_SeqAIJ,
2209: /* 4*/ MatMultAdd_SeqAIJ,
2210: MatMultTranspose_SeqAIJ,
2211: MatMultTransposeAdd_SeqAIJ,
2212: MatSolve_SeqAIJ,
2213: MatSolveAdd_SeqAIJ,
2214: MatSolveTranspose_SeqAIJ,
2215: /*10*/ MatSolveTransposeAdd_SeqAIJ,
2216: MatLUFactor_SeqAIJ,
2217: 0,
2218: MatRelax_SeqAIJ,
2219: MatTranspose_SeqAIJ,
2220: /*15*/ MatGetInfo_SeqAIJ,
2221: MatEqual_SeqAIJ,
2222: MatGetDiagonal_SeqAIJ,
2223: MatDiagonalScale_SeqAIJ,
2224: MatNorm_SeqAIJ,
2225: /*20*/ 0,
2226: MatAssemblyEnd_SeqAIJ,
2227: MatCompress_SeqAIJ,
2228: MatSetOption_SeqAIJ,
2229: MatZeroEntries_SeqAIJ,
2230: /*25*/ MatZeroRows_SeqAIJ,
2231: MatLUFactorSymbolic_SeqAIJ,
2232: MatLUFactorNumeric_SeqAIJ,
2233: MatCholeskyFactorSymbolic_SeqAIJ,
2234: MatCholeskyFactorNumeric_SeqAIJ,
2235: /*30*/ MatSetUpPreallocation_SeqAIJ,
2236: MatILUFactorSymbolic_SeqAIJ,
2237: MatICCFactorSymbolic_SeqAIJ,
2238: MatGetArray_SeqAIJ,
2239: MatRestoreArray_SeqAIJ,
2240: /*35*/ MatDuplicate_SeqAIJ,
2241: 0,
2242: 0,
2243: MatILUFactor_SeqAIJ,
2244: 0,
2245: /*40*/ MatAXPY_SeqAIJ,
2246: MatGetSubMatrices_SeqAIJ,
2247: MatIncreaseOverlap_SeqAIJ,
2248: MatGetValues_SeqAIJ,
2249: MatCopy_SeqAIJ,
2250: /*45*/ 0,
2251: MatScale_SeqAIJ,
2252: 0,
2253: MatDiagonalSet_SeqAIJ,
2254: MatILUDTFactor_SeqAIJ,
2255: /*50*/ MatSetBlockSize_SeqAIJ,
2256: MatGetRowIJ_SeqAIJ,
2257: MatRestoreRowIJ_SeqAIJ,
2258: MatGetColumnIJ_SeqAIJ,
2259: MatRestoreColumnIJ_SeqAIJ,
2260: /*55*/ MatFDColoringCreate_SeqAIJ,
2261: 0,
2262: 0,
2263: MatPermute_SeqAIJ,
2264: 0,
2265: /*60*/ 0,
2266: MatDestroy_SeqAIJ,
2267: MatView_SeqAIJ,
2268: 0,
2269: 0,
2270: /*65*/ 0,
2271: 0,
2272: 0,
2273: 0,
2274: 0,
2275: /*70*/ MatGetRowMax_SeqAIJ,
2276: 0,
2277: MatSetColoring_SeqAIJ,
2278: #if defined(PETSC_HAVE_ADIC)
2279: MatSetValuesAdic_SeqAIJ,
2280: #else
2281: 0,
2282: #endif
2283: MatSetValuesAdifor_SeqAIJ,
2284: /*75*/ 0,
2285: 0,
2286: 0,
2287: 0,
2288: 0,
2289: /*80*/ 0,
2290: 0,
2291: 0,
2292: 0,
2293: MatLoad_SeqAIJ,
2294: /*85*/ MatIsSymmetric_SeqAIJ,
2295: 0,
2296: 0,
2297: 0,
2298: 0,
2299: /*90*/ MatMatMult_SeqAIJ_SeqAIJ,
2300: MatMatMultSymbolic_SeqAIJ_SeqAIJ,
2301: MatMatMultNumeric_SeqAIJ_SeqAIJ,
2302: MatPtAP_Basic,
2303: MatPtAPSymbolic_SeqAIJ,
2304: /*95*/ MatPtAPNumeric_SeqAIJ,
2305: MatMatMultTranspose_SeqAIJ_SeqAIJ,
2306: MatMatMultTransposeSymbolic_SeqAIJ_SeqAIJ,
2307: MatMatMultTransposeNumeric_SeqAIJ_SeqAIJ,
2308: MatPtAPSymbolic_SeqAIJ_SeqAIJ,
2309: /*100*/MatPtAPNumeric_SeqAIJ_SeqAIJ,
2310: 0,
2311: 0,
2312: MatConjugate_SeqAIJ,
2313: 0,
2314: /*105*/MatSetValuesRow_SeqAIJ,
2315: MatRealPart_SeqAIJ,
2316: MatImaginaryPart_SeqAIJ,
2317: 0,
2318: 0,
2319: /*110*/MatMatSolve_SeqAIJ
2320: };
2325: PetscErrorCode MatSeqAIJSetColumnIndices_SeqAIJ(Mat mat,PetscInt *indices)
2326: {
2327: Mat_SeqAIJ *aij = (Mat_SeqAIJ *)mat->data;
2328: PetscInt i,nz,n;
2332: nz = aij->maxnz;
2333: n = mat->cmap.n;
2334: for (i=0; i<nz; i++) {
2335: aij->j[i] = indices[i];
2336: }
2337: aij->nz = nz;
2338: for (i=0; i<n; i++) {
2339: aij->ilen[i] = aij->imax[i];
2340: }
2342: return(0);
2343: }
2348: /*@
2349: MatSeqAIJSetColumnIndices - Set the column indices for all the rows
2350: in the matrix.
2352: Input Parameters:
2353: + mat - the SeqAIJ matrix
2354: - indices - the column indices
2356: Level: advanced
2358: Notes:
2359: This can be called if you have precomputed the nonzero structure of the
2360: matrix and want to provide it to the matrix object to improve the performance
2361: of the MatSetValues() operation.
2363: You MUST have set the correct numbers of nonzeros per row in the call to
2364: MatCreateSeqAIJ(), and the columns indices MUST be sorted.
2366: MUST be called before any calls to MatSetValues();
2368: The indices should start with zero, not one.
2370: @*/
2371: PetscErrorCode MatSeqAIJSetColumnIndices(Mat mat,PetscInt *indices)
2372: {
2373: PetscErrorCode ierr,(*f)(Mat,PetscInt *);
2378: PetscObjectQueryFunction((PetscObject)mat,"MatSeqAIJSetColumnIndices_C",(void (**)(void))&f);
2379: if (f) {
2380: (*f)(mat,indices);
2381: } else {
2382: SETERRQ(PETSC_ERR_SUP,"Wrong type of matrix to set column indices");
2383: }
2384: return(0);
2385: }
2387: /* ----------------------------------------------------------------------------------------*/
2392: PetscErrorCode MatStoreValues_SeqAIJ(Mat mat)
2393: {
2394: Mat_SeqAIJ *aij = (Mat_SeqAIJ *)mat->data;
2396: size_t nz = aij->i[mat->rmap.n];
2399: if (aij->nonew != 1) {
2400: SETERRQ(PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NO_NEW_NONZERO_LOCATIONS);first");
2401: }
2403: /* allocate space for values if not already there */
2404: if (!aij->saved_values) {
2405: PetscMalloc((nz+1)*sizeof(PetscScalar),&aij->saved_values);
2406: }
2408: /* copy values over */
2409: PetscMemcpy(aij->saved_values,aij->a,nz*sizeof(PetscScalar));
2410: return(0);
2411: }
2416: /*@
2417: MatStoreValues - Stashes a copy of the matrix values; this allows, for
2418: example, reuse of the linear part of a Jacobian, while recomputing the
2419: nonlinear portion.
2421: Collect on Mat
2423: Input Parameters:
2424: . mat - the matrix (currently only AIJ matrices support this option)
2426: Level: advanced
2428: Common Usage, with SNESSolve():
2429: $ Create Jacobian matrix
2430: $ Set linear terms into matrix
2431: $ Apply boundary conditions to matrix, at this time matrix must have
2432: $ final nonzero structure (i.e. setting the nonlinear terms and applying
2433: $ boundary conditions again will not change the nonzero structure
2434: $ MatSetOption(mat,MAT_NO_NEW_NONZERO_LOCATIONS);
2435: $ MatStoreValues(mat);
2436: $ Call SNESSetJacobian() with matrix
2437: $ In your Jacobian routine
2438: $ MatRetrieveValues(mat);
2439: $ Set nonlinear terms in matrix
2440:
2441: Common Usage without SNESSolve(), i.e. when you handle nonlinear solve yourself:
2442: $ // build linear portion of Jacobian
2443: $ MatSetOption(mat,MAT_NO_NEW_NONZERO_LOCATIONS);
2444: $ MatStoreValues(mat);
2445: $ loop over nonlinear iterations
2446: $ MatRetrieveValues(mat);
2447: $ // call MatSetValues(mat,...) to set nonliner portion of Jacobian
2448: $ // call MatAssemblyBegin/End() on matrix
2449: $ Solve linear system with Jacobian
2450: $ endloop
2452: Notes:
2453: Matrix must already be assemblied before calling this routine
2454: Must set the matrix option MatSetOption(mat,MAT_NO_NEW_NONZERO_LOCATIONS); before
2455: calling this routine.
2457: When this is called multiple times it overwrites the previous set of stored values
2458: and does not allocated additional space.
2460: .seealso: MatRetrieveValues()
2462: @*/
2463: PetscErrorCode MatStoreValues(Mat mat)
2464: {
2465: PetscErrorCode ierr,(*f)(Mat);
2469: if (!mat->assembled) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix");
2470: if (mat->factor) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
2472: PetscObjectQueryFunction((PetscObject)mat,"MatStoreValues_C",(void (**)(void))&f);
2473: if (f) {
2474: (*f)(mat);
2475: } else {
2476: SETERRQ(PETSC_ERR_SUP,"Wrong type of matrix to store values");
2477: }
2478: return(0);
2479: }
2484: PetscErrorCode MatRetrieveValues_SeqAIJ(Mat mat)
2485: {
2486: Mat_SeqAIJ *aij = (Mat_SeqAIJ *)mat->data;
2488: PetscInt nz = aij->i[mat->rmap.n];
2491: if (aij->nonew != 1) {
2492: SETERRQ(PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NO_NEW_NONZERO_LOCATIONS);first");
2493: }
2494: if (!aij->saved_values) {
2495: SETERRQ(PETSC_ERR_ORDER,"Must call MatStoreValues(A);first");
2496: }
2497: /* copy values over */
2498: PetscMemcpy(aij->a,aij->saved_values,nz*sizeof(PetscScalar));
2499: return(0);
2500: }
2505: /*@
2506: MatRetrieveValues - Retrieves the copy of the matrix values; this allows, for
2507: example, reuse of the linear part of a Jacobian, while recomputing the
2508: nonlinear portion.
2510: Collect on Mat
2512: Input Parameters:
2513: . mat - the matrix (currently on AIJ matrices support this option)
2515: Level: advanced
2517: .seealso: MatStoreValues()
2519: @*/
2520: PetscErrorCode MatRetrieveValues(Mat mat)
2521: {
2522: PetscErrorCode ierr,(*f)(Mat);
2526: if (!mat->assembled) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix");
2527: if (mat->factor) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
2529: PetscObjectQueryFunction((PetscObject)mat,"MatRetrieveValues_C",(void (**)(void))&f);
2530: if (f) {
2531: (*f)(mat);
2532: } else {
2533: SETERRQ(PETSC_ERR_SUP,"Wrong type of matrix to retrieve values");
2534: }
2535: return(0);
2536: }
2539: /* --------------------------------------------------------------------------------*/
2542: /*@C
2543: MatCreateSeqAIJ - Creates a sparse matrix in AIJ (compressed row) format
2544: (the default parallel PETSc format). For good matrix assembly performance
2545: the user should preallocate the matrix storage by setting the parameter nz
2546: (or the array nnz). By setting these parameters accurately, performance
2547: during matrix assembly can be increased by more than a factor of 50.
2549: Collective on MPI_Comm
2551: Input Parameters:
2552: + comm - MPI communicator, set to PETSC_COMM_SELF
2553: . m - number of rows
2554: . n - number of columns
2555: . nz - number of nonzeros per row (same for all rows)
2556: - nnz - array containing the number of nonzeros in the various rows
2557: (possibly different for each row) or PETSC_NULL
2559: Output Parameter:
2560: . A - the matrix
2562: Notes:
2563: If nnz is given then nz is ignored
2565: The AIJ format (also called the Yale sparse matrix format or
2566: compressed row storage), is fully compatible with standard Fortran 77
2567: storage. That is, the stored row and column indices can begin at
2568: either one (as in Fortran) or zero. See the users' manual for details.
2570: Specify the preallocated storage with either nz or nnz (not both).
2571: Set nz=PETSC_DEFAULT and nnz=PETSC_NULL for PETSc to control dynamic memory
2572: allocation. For large problems you MUST preallocate memory or you
2573: will get TERRIBLE performance, see the users' manual chapter on matrices.
2575: By default, this format uses inodes (identical nodes) when possible, to
2576: improve numerical efficiency of matrix-vector products and solves. We
2577: search for consecutive rows with the same nonzero structure, thereby
2578: reusing matrix information to achieve increased efficiency.
2580: Options Database Keys:
2581: + -mat_no_inode - Do not use inodes
2582: . -mat_inode_limit <limit> - Sets inode limit (max limit=5)
2583: - -mat_aij_oneindex - Internally use indexing starting at 1
2584: rather than 0. Note that when calling MatSetValues(),
2585: the user still MUST index entries starting at 0!
2587: Level: intermediate
2589: .seealso: MatCreate(), MatCreateMPIAIJ(), MatSetValues(), MatSeqAIJSetColumnIndices(), MatCreateSeqAIJWithArrays()
2591: @*/
2592: PetscErrorCode MatCreateSeqAIJ(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt nz,const PetscInt nnz[],Mat *A)
2593: {
2597: MatCreate(comm,A);
2598: MatSetSizes(*A,m,n,m,n);
2599: MatSetType(*A,MATSEQAIJ);
2600: MatSeqAIJSetPreallocation_SeqAIJ(*A,nz,(PetscInt*)nnz);
2601: return(0);
2602: }
2606: /*@C
2607: MatSeqAIJSetPreallocation - For good matrix assembly performance
2608: the user should preallocate the matrix storage by setting the parameter nz
2609: (or the array nnz). By setting these parameters accurately, performance
2610: during matrix assembly can be increased by more than a factor of 50.
2612: Collective on MPI_Comm
2614: Input Parameters:
2615: + B - The matrix
2616: . nz - number of nonzeros per row (same for all rows)
2617: - nnz - array containing the number of nonzeros in the various rows
2618: (possibly different for each row) or PETSC_NULL
2620: Notes:
2621: If nnz is given then nz is ignored
2623: The AIJ format (also called the Yale sparse matrix format or
2624: compressed row storage), is fully compatible with standard Fortran 77
2625: storage. That is, the stored row and column indices can begin at
2626: either one (as in Fortran) or zero. See the users' manual for details.
2628: Specify the preallocated storage with either nz or nnz (not both).
2629: Set nz=PETSC_DEFAULT and nnz=PETSC_NULL for PETSc to control dynamic memory
2630: allocation. For large problems you MUST preallocate memory or you
2631: will get TERRIBLE performance, see the users' manual chapter on matrices.
2633: Developers: Use nz of MAT_SKIP_ALLOCATION to not allocate any space for the matrix
2634: entries or columns indices
2636: By default, this format uses inodes (identical nodes) when possible, to
2637: improve numerical efficiency of matrix-vector products and solves. We
2638: search for consecutive rows with the same nonzero structure, thereby
2639: reusing matrix information to achieve increased efficiency.
2641: Options Database Keys:
2642: + -mat_no_inode - Do not use inodes
2643: . -mat_inode_limit <limit> - Sets inode limit (max limit=5)
2644: - -mat_aij_oneindex - Internally use indexing starting at 1
2645: rather than 0. Note that when calling MatSetValues(),
2646: the user still MUST index entries starting at 0!
2648: Level: intermediate
2650: .seealso: MatCreate(), MatCreateMPIAIJ(), MatSetValues(), MatSeqAIJSetColumnIndices(), MatCreateSeqAIJWithArrays()
2652: @*/
2653: PetscErrorCode MatSeqAIJSetPreallocation(Mat B,PetscInt nz,const PetscInt nnz[])
2654: {
2655: PetscErrorCode ierr,(*f)(Mat,PetscInt,const PetscInt[]);
2658: PetscObjectQueryFunction((PetscObject)B,"MatSeqAIJSetPreallocation_C",(void (**)(void))&f);
2659: if (f) {
2660: (*f)(B,nz,nnz);
2661: }
2662: return(0);
2663: }
2668: PetscErrorCode MatSeqAIJSetPreallocation_SeqAIJ(Mat B,PetscInt nz,PetscInt *nnz)
2669: {
2670: Mat_SeqAIJ *b;
2671: PetscTruth skipallocation = PETSC_FALSE;
2673: PetscInt i;
2676:
2677: if (nz == MAT_SKIP_ALLOCATION) {
2678: skipallocation = PETSC_TRUE;
2679: nz = 0;
2680: }
2682: B->rmap.bs = B->cmap.bs = 1;
2683: PetscMapInitialize(B->comm,&B->rmap);
2684: PetscMapInitialize(B->comm,&B->cmap);
2686: if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 5;
2687: if (nz < 0) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"nz cannot be less than 0: value %d",nz);
2688: if (nnz) {
2689: for (i=0; i<B->rmap.n; i++) {
2690: if (nnz[i] < 0) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"nnz cannot be less than 0: local row %d value %d",i,nnz[i]);
2691: if (nnz[i] > B->cmap.n) SETERRQ3(PETSC_ERR_ARG_OUTOFRANGE,"nnz cannot be greater than row length: local row %d value %d rowlength %d",i,nnz[i],B->cmap.n);
2692: }
2693: }
2695: B->preallocated = PETSC_TRUE;
2696: b = (Mat_SeqAIJ*)B->data;
2698: if (!skipallocation) {
2699: PetscMalloc2(B->rmap.n,PetscInt,&b->imax,B->rmap.n,PetscInt,&b->ilen);
2700: if (!nnz) {
2701: if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 10;
2702: else if (nz <= 0) nz = 1;
2703: for (i=0; i<B->rmap.n; i++) b->imax[i] = nz;
2704: nz = nz*B->rmap.n;
2705: } else {
2706: nz = 0;
2707: for (i=0; i<B->rmap.n; i++) {b->imax[i] = nnz[i]; nz += nnz[i];}
2708: }
2710: /* b->ilen will count nonzeros in each row so far. */
2711: for (i=0; i<B->rmap.n; i++) { b->ilen[i] = 0;}
2713: /* allocate the matrix space */
2714: PetscMalloc3(nz,PetscScalar,&b->a,nz,PetscInt,&b->j,B->rmap.n+1,PetscInt,&b->i);
2715: b->i[0] = 0;
2716: for (i=1; i<B->rmap.n+1; i++) {
2717: b->i[i] = b->i[i-1] + b->imax[i-1];
2718: }
2719: b->singlemalloc = PETSC_TRUE;
2720: b->free_a = PETSC_TRUE;
2721: b->free_ij = PETSC_TRUE;
2722: } else {
2723: b->free_a = PETSC_FALSE;
2724: b->free_ij = PETSC_FALSE;
2725: }
2727: b->nz = 0;
2728: b->maxnz = nz;
2729: B->info.nz_unneeded = (double)b->maxnz;
2730: return(0);
2731: }
2734: #undef __FUNCT__
2736: /*@C
2737: MatSeqAIJSetPreallocationCSR - Allocates memory for a sparse sequential matrix in AIJ format.
2739: Input Parameters:
2740: + B - the matrix
2741: . i - the indices into j for the start of each row (starts with zero)
2742: . j - the column indices for each row (starts with zero) these must be sorted for each row
2743: - v - optional values in the matrix
2745: Contributed by: Lisandro Dalchin
2747: Level: developer
2749: .keywords: matrix, aij, compressed row, sparse, sequential
2751: .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatSeqAIJSetPreallocation(), MatCreateSeqAIJ(), SeqAIJ
2752: @*/
2753: PetscErrorCode MatSeqAIJSetPreallocationCSR(Mat B,const PetscInt i[],const PetscInt j[],const PetscScalar v[])
2754: {
2755: PetscErrorCode (*f)(Mat,const PetscInt[],const PetscInt[],const PetscScalar[]);
2760: PetscObjectQueryFunction((PetscObject)B,"MatSeqAIJSetPreallocationCSR_C",(void (**)(void))&f);
2761: if (f) {
2762: (*f)(B,i,j,v);
2763: }
2764: return(0);
2765: }
2768: #undef __FUNCT__
2770: PetscErrorCode MatSeqAIJSetPreallocationCSR_SeqAIJ(Mat B,const PetscInt Ii[],const PetscInt J[],const PetscScalar v[])
2771: {
2772: PetscInt i;
2773: PetscInt m,n;
2774: PetscInt nz;
2775: PetscInt *nnz, nz_max = 0;
2776: PetscScalar *values;
2780: MatGetSize(B, &m, &n);
2782: if (Ii[0]) {
2783: SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE, "Ii[0] must be 0 it is %D", Ii[0]);
2784: }
2785: PetscMalloc((m+1) * sizeof(PetscInt), &nnz);
2786: for(i = 0; i < m; i++) {
2787: nz = Ii[i+1]- Ii[i];
2788: nz_max = PetscMax(nz_max, nz);
2789: if (nz < 0) {
2790: SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE, "Local row %D has a negative number of columns %D", i, nnz);
2791: }
2792: nnz[i] = nz;
2793: }
2794: MatSeqAIJSetPreallocation(B, 0, nnz);
2795: PetscFree(nnz);
2797: if (v) {
2798: values = (PetscScalar*) v;
2799: } else {
2800: PetscMalloc((nz_max+1)*sizeof(PetscScalar), &values);
2801: PetscMemzero(values, nz_max*sizeof(PetscScalar));
2802: }
2804: MatSetOption(B,MAT_COLUMNS_SORTED);
2806: for(i = 0; i < m; i++) {
2807: nz = Ii[i+1] - Ii[i];
2808: MatSetValues_SeqAIJ(B, 1, &i, nz, J+Ii[i], values + (v ? Ii[i] : 0), INSERT_VALUES);
2809: }
2811: MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);
2812: MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);
2813: MatSetOption(B,MAT_COLUMNS_UNSORTED);
2815: if (!v) {
2816: PetscFree(values);
2817: }
2818: return(0);
2819: }
2822: /*MC
2823: MATSEQAIJ - MATSEQAIJ = "seqaij" - A matrix type to be used for sequential sparse matrices,
2824: based on compressed sparse row format.
2826: Options Database Keys:
2827: . -mat_type seqaij - sets the matrix type to "seqaij" during a call to MatSetFromOptions()
2829: Level: beginner
2831: .seealso: MatCreateSeqAIJ(), MatSetFromOptions(), MatSetType(), MatCreate(), MatType
2832: M*/
2841: PetscErrorCode MatCreate_SeqAIJ(Mat B)
2842: {
2843: Mat_SeqAIJ *b;
2845: PetscMPIInt size;
2848: MPI_Comm_size(B->comm,&size);
2849: if (size > 1) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Comm must be of size 1");
2851: PetscNew(Mat_SeqAIJ,&b);
2852: B->data = (void*)b;
2853: PetscMemcpy(B->ops,&MatOps_Values,sizeof(struct _MatOps));
2854: B->factor = 0;
2855: B->mapping = 0;
2856: b->row = 0;
2857: b->col = 0;
2858: b->icol = 0;
2859: b->reallocs = 0;
2860: b->sorted = PETSC_FALSE;
2861: b->ignorezeroentries = PETSC_FALSE;
2862: b->roworiented = PETSC_TRUE;
2863: b->nonew = 0;
2864: b->diag = 0;
2865: b->solve_work = 0;
2866: B->spptr = 0;
2867: b->saved_values = 0;
2868: b->idiag = 0;
2869: b->ssor = 0;
2870: b->keepzeroedrows = PETSC_FALSE;
2871: b->xtoy = 0;
2872: b->XtoY = 0;
2873: b->compressedrow.use = PETSC_FALSE;
2874: b->compressedrow.nrows = B->rmap.n;
2875: b->compressedrow.i = PETSC_NULL;
2876: b->compressedrow.rindex = PETSC_NULL;
2877: b->compressedrow.checked = PETSC_FALSE;
2878: B->same_nonzero = PETSC_FALSE;
2880: PetscObjectChangeTypeName((PetscObject)B,MATSEQAIJ);
2881: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatSeqAIJSetColumnIndices_C",
2882: "MatSeqAIJSetColumnIndices_SeqAIJ",
2883: MatSeqAIJSetColumnIndices_SeqAIJ);
2884: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatStoreValues_C",
2885: "MatStoreValues_SeqAIJ",
2886: MatStoreValues_SeqAIJ);
2887: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatRetrieveValues_C",
2888: "MatRetrieveValues_SeqAIJ",
2889: MatRetrieveValues_SeqAIJ);
2890: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatConvert_seqaij_seqsbaij_C",
2891: "MatConvert_SeqAIJ_SeqSBAIJ",
2892: MatConvert_SeqAIJ_SeqSBAIJ);
2893: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatConvert_seqaij_seqbaij_C",
2894: "MatConvert_SeqAIJ_SeqBAIJ",
2895: MatConvert_SeqAIJ_SeqBAIJ);
2896: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatConvert_seqaij_seqcsrperm_C",
2897: "MatConvert_SeqAIJ_SeqCSRPERM",
2898: MatConvert_SeqAIJ_SeqCSRPERM);
2899: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatConvert_seqaij_seqcrl_C",
2900: "MatConvert_SeqAIJ_SeqCRL",
2901: MatConvert_SeqAIJ_SeqCRL);
2902: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatIsTranspose_C",
2903: "MatIsTranspose_SeqAIJ",
2904: MatIsTranspose_SeqAIJ);
2905: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatSeqAIJSetPreallocation_C",
2906: "MatSeqAIJSetPreallocation_SeqAIJ",
2907: MatSeqAIJSetPreallocation_SeqAIJ);
2908: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatSeqAIJSetPreallocationCSR_C",
2909: "MatSeqAIJSetPreallocationCSR_SeqAIJ",
2910: MatSeqAIJSetPreallocationCSR_SeqAIJ);
2911: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatReorderForNonzeroDiagonal_C",
2912: "MatReorderForNonzeroDiagonal_SeqAIJ",
2913: MatReorderForNonzeroDiagonal_SeqAIJ);
2914: MatCreate_Inode(B);
2915: PetscObjectChangeTypeName((PetscObject)B,MATSEQAIJ);
2916: return(0);
2917: }
2922: PetscErrorCode MatDuplicate_SeqAIJ(Mat A,MatDuplicateOption cpvalues,Mat *B)
2923: {
2924: Mat C;
2925: Mat_SeqAIJ *c,*a = (Mat_SeqAIJ*)A->data;
2927: PetscInt i,m = A->rmap.n;
2930: *B = 0;
2931: MatCreate(A->comm,&C);
2932: MatSetSizes(C,A->rmap.n,A->cmap.n,A->rmap.n,A->cmap.n);
2933: MatSetType(C,A->type_name);
2934: PetscMemcpy(C->ops,A->ops,sizeof(struct _MatOps));
2935:
2936: PetscMapCopy(A->comm,&A->rmap,&C->rmap);
2937: PetscMapCopy(A->comm,&A->cmap,&C->cmap);
2939: c = (Mat_SeqAIJ*)C->data;
2941: C->factor = A->factor;
2943: c->row = 0;
2944: c->col = 0;
2945: c->icol = 0;
2946: c->reallocs = 0;
2948: C->assembled = PETSC_TRUE;
2949:
2950: PetscMalloc2(m,PetscInt,&c->imax,m,PetscInt,&c->ilen);
2951: for (i=0; i<m; i++) {
2952: c->imax[i] = a->imax[i];
2953: c->ilen[i] = a->ilen[i];
2954: }
2956: /* allocate the matrix space */
2957: PetscMalloc3(a->i[m],PetscScalar,&c->a,a->i[m],PetscInt,&c->j,m+1,PetscInt,&c->i);
2958: c->singlemalloc = PETSC_TRUE;
2959: PetscMemcpy(c->i,a->i,(m+1)*sizeof(PetscInt));
2960: if (m > 0) {
2961: PetscMemcpy(c->j,a->j,(a->i[m])*sizeof(PetscInt));
2962: if (cpvalues == MAT_COPY_VALUES) {
2963: PetscMemcpy(c->a,a->a,(a->i[m])*sizeof(PetscScalar));
2964: } else {
2965: PetscMemzero(c->a,(a->i[m])*sizeof(PetscScalar));
2966: }
2967: }
2969: c->sorted = a->sorted;
2970: c->ignorezeroentries = a->ignorezeroentries;
2971: c->roworiented = a->roworiented;
2972: c->nonew = a->nonew;
2973: if (a->diag) {
2974: PetscMalloc((m+1)*sizeof(PetscInt),&c->diag);
2975: PetscLogObjectMemory(C,(m+1)*sizeof(PetscInt));
2976: for (i=0; i<m; i++) {
2977: c->diag[i] = a->diag[i];
2978: }
2979: } else c->diag = 0;
2980: c->solve_work = 0;
2981: c->saved_values = 0;
2982: c->idiag = 0;
2983: c->ssor = 0;
2984: c->keepzeroedrows = a->keepzeroedrows;
2985: c->free_a = PETSC_TRUE;
2986: c->free_ij = PETSC_TRUE;
2987: c->xtoy = 0;
2988: c->XtoY = 0;
2990: c->nz = a->nz;
2991: c->maxnz = a->maxnz;
2992: C->preallocated = PETSC_TRUE;
2994: c->compressedrow.use = a->compressedrow.use;
2995: c->compressedrow.nrows = a->compressedrow.nrows;
2996: c->compressedrow.checked = a->compressedrow.checked;
2997: if ( a->compressedrow.checked && a->compressedrow.use){
2998: i = a->compressedrow.nrows;
2999: PetscMalloc((2*i+1)*sizeof(PetscInt),&c->compressedrow.i);
3000: c->compressedrow.rindex = c->compressedrow.i + i + 1;
3001: PetscMemcpy(c->compressedrow.i,a->compressedrow.i,(i+1)*sizeof(PetscInt));
3002: PetscMemcpy(c->compressedrow.rindex,a->compressedrow.rindex,i*sizeof(PetscInt));
3003: } else {
3004: c->compressedrow.use = PETSC_FALSE;
3005: c->compressedrow.i = PETSC_NULL;
3006: c->compressedrow.rindex = PETSC_NULL;
3007: }
3008: C->same_nonzero = A->same_nonzero;
3009: MatDuplicate_Inode(A,cpvalues,&C);
3011: *B = C;
3012: PetscFListDuplicate(A->qlist,&C->qlist);
3013: return(0);
3014: }
3018: PetscErrorCode MatLoad_SeqAIJ(PetscViewer viewer, MatType type,Mat *A)
3019: {
3020: Mat_SeqAIJ *a;
3021: Mat B;
3023: PetscInt i,sum,nz,header[4],*rowlengths = 0,M,N;
3024: int fd;
3025: PetscMPIInt size;
3026: MPI_Comm comm;
3027:
3029: PetscObjectGetComm((PetscObject)viewer,&comm);
3030: MPI_Comm_size(comm,&size);
3031: if (size > 1) SETERRQ(PETSC_ERR_ARG_SIZ,"view must have one processor");
3032: PetscViewerBinaryGetDescriptor(viewer,&fd);
3033: PetscBinaryRead(fd,header,4,PETSC_INT);
3034: if (header[0] != MAT_FILE_COOKIE) SETERRQ(PETSC_ERR_FILE_UNEXPECTED,"not matrix object in file");
3035: M = header[1]; N = header[2]; nz = header[3];
3037: if (nz < 0) {
3038: SETERRQ(PETSC_ERR_FILE_UNEXPECTED,"Matrix stored in special format on disk,cannot load as SeqAIJ");
3039: }
3041: /* read in row lengths */
3042: PetscMalloc(M*sizeof(PetscInt),&rowlengths);
3043: PetscBinaryRead(fd,rowlengths,M,PETSC_INT);
3045: /* check if sum of rowlengths is same as nz */
3046: for (i=0,sum=0; i< M; i++) sum +=rowlengths[i];
3047: if (sum != nz) SETERRQ2(PETSC_ERR_FILE_READ,"Inconsistant matrix data in file. no-nonzeros = %d, sum-row-lengths = %d\n",nz,sum);
3049: /* create our matrix */
3050: MatCreate(comm,&B);
3051: MatSetSizes(B,PETSC_DECIDE,PETSC_DECIDE,M,N);
3052: MatSetType(B,type);
3053: MatSeqAIJSetPreallocation_SeqAIJ(B,0,rowlengths);
3054: a = (Mat_SeqAIJ*)B->data;
3056: PetscBinaryRead(fd,a->j,nz,PETSC_INT);
3058: /* read in nonzero values */
3059: PetscBinaryRead(fd,a->a,nz,PETSC_SCALAR);
3061: /* set matrix "i" values */
3062: a->i[0] = 0;
3063: for (i=1; i<= M; i++) {
3064: a->i[i] = a->i[i-1] + rowlengths[i-1];
3065: a->ilen[i-1] = rowlengths[i-1];
3066: }
3067: PetscFree(rowlengths);
3069: MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);
3070: MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);
3071: *A = B;
3072: return(0);
3073: }
3077: PetscErrorCode MatEqual_SeqAIJ(Mat A,Mat B,PetscTruth* flg)
3078: {
3079: Mat_SeqAIJ *a = (Mat_SeqAIJ *)A->data,*b = (Mat_SeqAIJ *)B->data;
3083: /* If the matrix dimensions are not equal,or no of nonzeros */
3084: if ((A->rmap.n != B->rmap.n) || (A->cmap.n != B->cmap.n) ||(a->nz != b->nz)) {
3085: *flg = PETSC_FALSE;
3086: return(0);
3087: }
3088:
3089: /* if the a->i are the same */
3090: PetscMemcmp(a->i,b->i,(A->rmap.n+1)*sizeof(PetscInt),flg);
3091: if (!*flg) return(0);
3092:
3093: /* if a->j are the same */
3094: PetscMemcmp(a->j,b->j,(a->nz)*sizeof(PetscInt),flg);
3095: if (!*flg) return(0);
3096:
3097: /* if a->a are the same */
3098: PetscMemcmp(a->a,b->a,(a->nz)*sizeof(PetscScalar),flg);
3100: return(0);
3101:
3102: }
3106: /*@
3107: MatCreateSeqAIJWithArrays - Creates an sequential AIJ matrix using matrix elements (in CSR format)
3108: provided by the user.
3110: Collective on MPI_Comm
3112: Input Parameters:
3113: + comm - must be an MPI communicator of size 1
3114: . m - number of rows
3115: . n - number of columns
3116: . i - row indices
3117: . j - column indices
3118: - a - matrix values
3120: Output Parameter:
3121: . mat - the matrix
3123: Level: intermediate
3125: Notes:
3126: The i, j, and a arrays are not copied by this routine, the user must free these arrays
3127: once the matrix is destroyed
3129: You cannot set new nonzero locations into this matrix, that will generate an error.
3131: The i and j indices are 0 based
3133: .seealso: MatCreate(), MatCreateMPIAIJ(), MatCreateSeqAIJ(), MatCreateMPIAIJWithArrays(), MatMPIAIJSetPreallocationCSR()
3135: @*/
3136: PetscErrorCode MatCreateSeqAIJWithArrays(MPI_Comm comm,PetscInt m,PetscInt n,PetscInt* i,PetscInt*j,PetscScalar *a,Mat *mat)
3137: {
3139: PetscInt ii;
3140: Mat_SeqAIJ *aij;
3143: if (i[0]) {
3144: SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"i (row indices) must start with 0");
3145: }
3146: MatCreate(comm,mat);
3147: MatSetSizes(*mat,m,n,m,n);
3148: MatSetType(*mat,MATSEQAIJ);
3149: MatSeqAIJSetPreallocation_SeqAIJ(*mat,MAT_SKIP_ALLOCATION,0);
3150: aij = (Mat_SeqAIJ*)(*mat)->data;
3151: PetscMalloc2(m,PetscInt,&aij->imax,m,PetscInt,&aij->ilen);
3153: aij->i = i;
3154: aij->j = j;
3155: aij->a = a;
3156: aij->singlemalloc = PETSC_FALSE;
3157: aij->nonew = -1; /*this indicates that inserting a new value in the matrix that generates a new nonzero is an error*/
3158: aij->free_a = PETSC_FALSE;
3159: aij->free_ij = PETSC_FALSE;
3161: for (ii=0; ii<m; ii++) {
3162: aij->ilen[ii] = aij->imax[ii] = i[ii+1] - i[ii];
3163: #if defined(PETSC_USE_DEBUG)
3164: if (i[ii+1] - i[ii] < 0) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Negative row length in i (row indices) row = %d length = %d",ii,i[ii+1] - i[ii]);
3165: #endif
3166: }
3167: #if defined(PETSC_USE_DEBUG)
3168: for (ii=0; ii<aij->i[m]; ii++) {
3169: if (j[ii] < 0) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Negative column index at location = %d index = %d",ii,j[ii]);
3170: if (j[ii] > n - 1) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Column index to large at location = %d index = %d",ii,j[ii]);
3171: }
3172: #endif
3174: MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);
3175: MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);
3176: return(0);
3177: }
3181: PetscErrorCode MatSetColoring_SeqAIJ(Mat A,ISColoring coloring)
3182: {
3184: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
3187: if (coloring->ctype == IS_COLORING_GLOBAL) {
3188: ISColoringReference(coloring);
3189: a->coloring = coloring;
3190: } else if (coloring->ctype == IS_COLORING_GHOSTED) {
3191: PetscInt i,*larray;
3192: ISColoring ocoloring;
3193: ISColoringValue *colors;
3195: /* set coloring for diagonal portion */
3196: PetscMalloc((A->cmap.n+1)*sizeof(PetscInt),&larray);
3197: for (i=0; i<A->cmap.n; i++) {
3198: larray[i] = i;
3199: }
3200: ISGlobalToLocalMappingApply(A->mapping,IS_GTOLM_MASK,A->cmap.n,larray,PETSC_NULL,larray);
3201: PetscMalloc((A->cmap.n+1)*sizeof(ISColoringValue),&colors);
3202: for (i=0; i<A->cmap.n; i++) {
3203: colors[i] = coloring->colors[larray[i]];
3204: }
3205: PetscFree(larray);
3206: ISColoringCreate(PETSC_COMM_SELF,coloring->n,A->cmap.n,colors,&ocoloring);
3207: a->coloring = ocoloring;
3208: }
3209: return(0);
3210: }
3212: #if defined(PETSC_HAVE_ADIC)
3214: #include "adic/ad_utils.h"
3219: PetscErrorCode MatSetValuesAdic_SeqAIJ(Mat A,void *advalues)
3220: {
3221: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
3222: PetscInt m = A->rmap.n,*ii = a->i,*jj = a->j,nz,i,j,nlen;
3223: PetscScalar *v = a->a,*values = ((PetscScalar*)advalues)+1;
3224: ISColoringValue *color;
3227: if (!a->coloring) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Coloring not set for matrix");
3228: nlen = PetscADGetDerivTypeSize()/sizeof(PetscScalar);
3229: color = a->coloring->colors;
3230: /* loop over rows */
3231: for (i=0; i<m; i++) {
3232: nz = ii[i+1] - ii[i];
3233: /* loop over columns putting computed value into matrix */
3234: for (j=0; j<nz; j++) {
3235: *v++ = values[color[*jj++]];
3236: }
3237: values += nlen; /* jump to next row of derivatives */
3238: }
3239: return(0);
3240: }
3241: #endif
3245: PetscErrorCode MatSetValuesAdifor_SeqAIJ(Mat A,PetscInt nl,void *advalues)
3246: {
3247: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
3248: PetscInt m = A->rmap.n,*ii = a->i,*jj = a->j,nz,i,j;
3249: PetscScalar *v = a->a,*values = (PetscScalar *)advalues;
3250: ISColoringValue *color;
3253: if (!a->coloring) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Coloring not set for matrix");
3254: color = a->coloring->colors;
3255: /* loop over rows */
3256: for (i=0; i<m; i++) {
3257: nz = ii[i+1] - ii[i];
3258: /* loop over columns putting computed value into matrix */
3259: for (j=0; j<nz; j++) {
3260: *v++ = values[color[*jj++]];
3261: }
3262: values += nl; /* jump to next row of derivatives */
3263: }
3264: return(0);
3265: }
3267: /*
3268: Special version for direct calls from Fortran
3269: */
3270: #if defined(PETSC_HAVE_FORTRAN_CAPS)
3271: #define matsetvaluesseqaij_ MATSETVALUESSEQAIJ
3272: #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE)
3273: #define matsetvaluesseqaij_ matsetvaluesseqaij
3274: #endif
3276: /* Change these macros so can be used in void function */
3277: #undef CHKERRQ
3278: #define CHKERRQ(ierr) CHKERRABORT(A->comm,ierr)
3279: #undef SETERRQ2
3280: #define SETERRQ2(ierr,b,c,d) CHKERRABORT(A->comm,ierr)
3285: void PETSC_STDCALL matsetvaluesseqaij_(Mat *AA,PetscInt *mm,const PetscInt im[],PetscInt *nn,const PetscInt in[],const PetscScalar v[],InsertMode *isis, PetscErrorCode *_ierr)
3286: {
3287: Mat A = *AA;
3288: PetscInt m = *mm, n = *nn;
3289: InsertMode is = *isis;
3290: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
3291: PetscInt *rp,k,low,high,t,ii,row,nrow,i,col,l,rmax,N;
3292: PetscInt *imax,*ai,*ailen;
3294: PetscInt *aj,nonew = a->nonew,lastcol = -1;
3295: PetscScalar *ap,value,*aa;
3296: PetscTruth ignorezeroentries = ((a->ignorezeroentries && is == ADD_VALUES) ? PETSC_TRUE:PETSC_FALSE);
3297: PetscTruth roworiented = a->roworiented;
3300: MatPreallocated(A);
3301: imax = a->imax;
3302: ai = a->i;
3303: ailen = a->ilen;
3304: aj = a->j;
3305: aa = a->a;
3307: for (k=0; k<m; k++) { /* loop over added rows */
3308: row = im[k];
3309: if (row < 0) continue;
3310: #if defined(PETSC_USE_DEBUG)
3311: if (row >= A->rmap.n) SETERRABORT(A->comm,PETSC_ERR_ARG_OUTOFRANGE,"Row too large");
3312: #endif
3313: rp = aj + ai[row]; ap = aa + ai[row];
3314: rmax = imax[row]; nrow = ailen[row];
3315: low = 0;
3316: high = nrow;
3317: for (l=0; l<n; l++) { /* loop over added columns */
3318: if (in[l] < 0) continue;
3319: #if defined(PETSC_USE_DEBUG)
3320: if (in[l] >= A->cmap.n) SETERRABORT(A->comm,PETSC_ERR_ARG_OUTOFRANGE,"Column too large");
3321: #endif
3322: col = in[l];
3323: if (roworiented) {
3324: value = v[l + k*n];
3325: } else {
3326: value = v[k + l*m];
3327: }
3328: if (value == 0.0 && ignorezeroentries && (is == ADD_VALUES)) continue;
3330: if (col <= lastcol) low = 0; else high = nrow;
3331: lastcol = col;
3332: while (high-low > 5) {
3333: t = (low+high)/2;
3334: if (rp[t] > col) high = t;
3335: else low = t;
3336: }
3337: for (i=low; i<high; i++) {
3338: if (rp[i] > col) break;
3339: if (rp[i] == col) {
3340: if (is == ADD_VALUES) ap[i] += value;
3341: else ap[i] = value;
3342: goto noinsert;
3343: }
3344: }
3345: if (value == 0.0 && ignorezeroentries) goto noinsert;
3346: if (nonew == 1) goto noinsert;
3347: if (nonew == -1) SETERRABORT(A->comm,PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero in the matrix");
3348: MatSeqXAIJReallocateAIJ(A,A->rmap.n,1,nrow,row,col,rmax,aa,ai,aj,rp,ap,imax,nonew,MatScalar);
3349: N = nrow++ - 1; a->nz++; high++;
3350: /* shift up all the later entries in this row */
3351: for (ii=N; ii>=i; ii--) {
3352: rp[ii+1] = rp[ii];
3353: ap[ii+1] = ap[ii];
3354: }
3355: rp[i] = col;
3356: ap[i] = value;
3357: noinsert:;
3358: low = i + 1;
3359: }
3360: ailen[row] = nrow;
3361: }
3362: A->same_nonzero = PETSC_FALSE;
3363: PetscFunctionReturnVoid();
3364: }