Actual source code: aij.c
1: /*$Id: aij.c,v 1.385 2001/09/07 20:09:22 bsmith Exp $*/
2: /*
3: Defines the basic matrix operations for the AIJ (compressed row)
4: matrix storage format.
5: */
7: #include src/mat/impls/aij/seq/aij.h
8: #include src/inline/spops.h
9: #include src/inline/dot.h
10: #include petscbt.h
14: int MatGetRowIJ_SeqAIJ(Mat A,int oshift,PetscTruth symmetric,int *m,int *ia[],int *ja[],PetscTruth *done)
15: {
16: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
17: int ierr,i,ishift;
18:
20: *m = A->m;
21: if (!ia) return(0);
22: ishift = 0;
23: if (symmetric && !A->structurally_symmetric) {
24: MatToSymmetricIJ_SeqAIJ(A->m,a->i,a->j,ishift,oshift,ia,ja);
25: } else if (oshift == 1) {
26: int nz = a->i[A->m];
27: /* malloc space and add 1 to i and j indices */
28: PetscMalloc((A->m+1)*sizeof(int),ia);
29: PetscMalloc((nz+1)*sizeof(int),ja);
30: for (i=0; i<nz; i++) (*ja)[i] = a->j[i] + 1;
31: for (i=0; i<A->m+1; i++) (*ia)[i] = a->i[i] + 1;
32: } else {
33: *ia = a->i; *ja = a->j;
34: }
35: return(0);
36: }
40: int MatRestoreRowIJ_SeqAIJ(Mat A,int oshift,PetscTruth symmetric,int *n,int *ia[],int *ja[],PetscTruth *done)
41: {
43:
45: if (!ia) return(0);
46: if ((symmetric && !A->structurally_symmetric) || oshift == 1) {
47: PetscFree(*ia);
48: PetscFree(*ja);
49: }
50: return(0);
51: }
55: int MatGetColumnIJ_SeqAIJ(Mat A,int oshift,PetscTruth symmetric,int *nn,int *ia[],int *ja[],PetscTruth *done)
56: {
57: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
58: int ierr,i,*collengths,*cia,*cja,n = A->n,m = A->m;
59: int nz = a->i[m],row,*jj,mr,col;
60:
62: *nn = A->n;
63: if (!ia) return(0);
64: if (symmetric) {
65: MatToSymmetricIJ_SeqAIJ(A->m,a->i,a->j,0,oshift,ia,ja);
66: } else {
67: PetscMalloc((n+1)*sizeof(int),&collengths);
68: PetscMemzero(collengths,n*sizeof(int));
69: PetscMalloc((n+1)*sizeof(int),&cia);
70: PetscMalloc((nz+1)*sizeof(int),&cja);
71: jj = a->j;
72: for (i=0; i<nz; i++) {
73: collengths[jj[i]]++;
74: }
75: cia[0] = oshift;
76: for (i=0; i<n; i++) {
77: cia[i+1] = cia[i] + collengths[i];
78: }
79: PetscMemzero(collengths,n*sizeof(int));
80: jj = a->j;
81: for (row=0; row<m; row++) {
82: mr = a->i[row+1] - a->i[row];
83: for (i=0; i<mr; i++) {
84: col = *jj++;
85: cja[cia[col] + collengths[col]++ - oshift] = row + oshift;
86: }
87: }
88: PetscFree(collengths);
89: *ia = cia; *ja = cja;
90: }
91: return(0);
92: }
96: int MatRestoreColumnIJ_SeqAIJ(Mat A,int oshift,PetscTruth symmetric,int *n,int *ia[],int *ja[],PetscTruth *done)
97: {
101: if (!ia) return(0);
103: PetscFree(*ia);
104: PetscFree(*ja);
105:
106: return(0);
107: }
109: #define CHUNKSIZE 15
113: int MatSetValues_SeqAIJ(Mat A,int m,const int im[],int n,const int in[],const PetscScalar v[],InsertMode is)
114: {
115: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
116: int *rp,k,low,high,t,ii,row,nrow,i,col,l,rmax,N,sorted = a->sorted;
117: int *imax = a->imax,*ai = a->i,*ailen = a->ilen;
118: int *aj = a->j,nonew = a->nonew,ierr;
119: PetscScalar *ap,value,*aa = a->a;
120: PetscTruth ignorezeroentries = ((a->ignorezeroentries && is == ADD_VALUES) ? PETSC_TRUE:PETSC_FALSE);
121: PetscTruth roworiented = a->roworiented;
124: for (k=0; k<m; k++) { /* loop over added rows */
125: row = im[k];
126: if (row < 0) continue;
127: #if defined(PETSC_USE_BOPT_g)
128: if (row >= A->m) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %d max %d",row,A->m-1);
129: #endif
130: rp = aj + ai[row]; ap = aa + ai[row];
131: rmax = imax[row]; nrow = ailen[row];
132: low = 0;
133: for (l=0; l<n; l++) { /* loop over added columns */
134: if (in[l] < 0) continue;
135: #if defined(PETSC_USE_BOPT_g)
136: if (in[l] >= A->n) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %d max %d",in[l],A->n-1);
137: #endif
138: col = in[l];
139: if (roworiented) {
140: value = v[l + k*n];
141: } else {
142: value = v[k + l*m];
143: }
144: if (value == 0.0 && ignorezeroentries) continue;
146: if (!sorted) low = 0; high = nrow;
147: while (high-low > 5) {
148: t = (low+high)/2;
149: if (rp[t] > col) high = t;
150: else low = t;
151: }
152: for (i=low; i<high; i++) {
153: if (rp[i] > col) break;
154: if (rp[i] == col) {
155: if (is == ADD_VALUES) ap[i] += value;
156: else ap[i] = value;
157: goto noinsert;
158: }
159: }
160: if (nonew == 1) goto noinsert;
161: else if (nonew == -1) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero at (%d,%d) in the matrix",row,col);
162: if (nrow >= rmax) {
163: /* there is no extra room in row, therefore enlarge */
164: int new_nz = ai[A->m] + CHUNKSIZE,*new_i,*new_j;
165: size_t len;
166: PetscScalar *new_a;
168: if (nonew == -2) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero at (%d,%d) in the matrix requiring new malloc()",row,col);
170: /* malloc new storage space */
171: len = ((size_t) new_nz)*(sizeof(int)+sizeof(PetscScalar))+(A->m+1)*sizeof(int);
172: PetscMalloc(len,&new_a);
173: new_j = (int*)(new_a + new_nz);
174: new_i = new_j + new_nz;
176: /* copy over old data into new slots */
177: for (ii=0; ii<row+1; ii++) {new_i[ii] = ai[ii];}
178: for (ii=row+1; ii<A->m+1; ii++) {new_i[ii] = ai[ii]+CHUNKSIZE;}
179: PetscMemcpy(new_j,aj,(ai[row]+nrow)*sizeof(int));
180: len = (((size_t) new_nz) - CHUNKSIZE - ai[row] - nrow );
181: PetscMemcpy(new_j+ai[row]+nrow+CHUNKSIZE,aj+ai[row]+nrow,len*sizeof(int));
182: PetscMemcpy(new_a,aa,(((size_t) ai[row])+nrow)*sizeof(PetscScalar));
183: PetscMemcpy(new_a+ai[row]+nrow+CHUNKSIZE,aa+ai[row]+nrow,len*sizeof(PetscScalar));
184: /* free up old matrix storage */
185: PetscFree(a->a);
186: if (!a->singlemalloc) {
187: PetscFree(a->i);
188: PetscFree(a->j);
189: }
190: aa = a->a = new_a; ai = a->i = new_i; aj = a->j = new_j;
191: a->singlemalloc = PETSC_TRUE;
193: rp = aj + ai[row]; ap = aa + ai[row] ;
194: rmax = imax[row] = imax[row] + CHUNKSIZE;
195: PetscLogObjectMemory(A,CHUNKSIZE*(sizeof(int) + sizeof(PetscScalar)));
196: a->maxnz += CHUNKSIZE;
197: a->reallocs++;
198: }
199: N = nrow++ - 1; a->nz++;
200: /* shift up all the later entries in this row */
201: for (ii=N; ii>=i; ii--) {
202: rp[ii+1] = rp[ii];
203: ap[ii+1] = ap[ii];
204: }
205: rp[i] = col;
206: ap[i] = value;
207: noinsert:;
208: low = i + 1;
209: }
210: ailen[row] = nrow;
211: }
212: return(0);
213: }
217: int MatGetValues_SeqAIJ(Mat A,int m,const int im[],int n,const int in[],PetscScalar v[])
218: {
219: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
220: int *rp,k,low,high,t,row,nrow,i,col,l,*aj = a->j;
221: int *ai = a->i,*ailen = a->ilen;
222: PetscScalar *ap,*aa = a->a,zero = 0.0;
225: for (k=0; k<m; k++) { /* loop over rows */
226: row = im[k];
227: if (row < 0) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"Negative row: %d",row);
228: if (row >= A->m) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %d max %d",row,A->m-1);
229: rp = aj + ai[row]; ap = aa + ai[row];
230: nrow = ailen[row];
231: for (l=0; l<n; l++) { /* loop over columns */
232: if (in[l] < 0) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"Negative column: %d",in[l]);
233: if (in[l] >= A->n) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %d max %d",in[l],A->n-1);
234: col = in[l] ;
235: high = nrow; low = 0; /* assume unsorted */
236: while (high-low > 5) {
237: t = (low+high)/2;
238: if (rp[t] > col) high = t;
239: else low = t;
240: }
241: for (i=low; i<high; i++) {
242: if (rp[i] > col) break;
243: if (rp[i] == col) {
244: *v++ = ap[i];
245: goto finished;
246: }
247: }
248: *v++ = zero;
249: finished:;
250: }
251: }
252: return(0);
253: }
258: int MatView_SeqAIJ_Binary(Mat A,PetscViewer viewer)
259: {
260: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
261: int i,fd,*col_lens,ierr;
264: PetscViewerBinaryGetDescriptor(viewer,&fd);
265: PetscMalloc((4+A->m)*sizeof(int),&col_lens);
266: col_lens[0] = MAT_FILE_COOKIE;
267: col_lens[1] = A->m;
268: col_lens[2] = A->n;
269: col_lens[3] = a->nz;
271: /* store lengths of each row and write (including header) to file */
272: for (i=0; i<A->m; i++) {
273: col_lens[4+i] = a->i[i+1] - a->i[i];
274: }
275: PetscBinaryWrite(fd,col_lens,4+A->m,PETSC_INT,1);
276: PetscFree(col_lens);
278: /* store column indices (zero start index) */
279: PetscBinaryWrite(fd,a->j,a->nz,PETSC_INT,0);
281: /* store nonzero values */
282: PetscBinaryWrite(fd,a->a,a->nz,PETSC_SCALAR,0);
283: return(0);
284: }
286: extern int MatSeqAIJFactorInfo_Matlab(Mat,PetscViewer);
290: int MatView_SeqAIJ_ASCII(Mat A,PetscViewer viewer)
291: {
292: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
293: int ierr,i,j,m = A->m,shift=0;
294: char *name;
295: PetscViewerFormat format;
298: PetscObjectGetName((PetscObject)A,&name);
299: PetscViewerGetFormat(viewer,&format);
300: if (format == PETSC_VIEWER_ASCII_INFO_DETAIL || format == PETSC_VIEWER_ASCII_INFO) {
301: if (a->inode.size) {
302: PetscViewerASCIIPrintf(viewer,"using I-node routines: found %d nodes, limit used is %d\n",a->inode.node_count,a->inode.limit);
303: } else {
304: PetscViewerASCIIPrintf(viewer,"not using I-node routines\n");
305: }
306: } else if (format == PETSC_VIEWER_ASCII_MATLAB) {
307: int nofinalvalue = 0;
308: if ((a->i[m] == a->i[m-1]) || (a->j[a->nz-1] != A->n-!shift)) {
309: nofinalvalue = 1;
310: }
311: PetscViewerASCIIUseTabs(viewer,PETSC_NO);
312: PetscViewerASCIIPrintf(viewer,"%% Size = %d %d \n",m,A->n);
313: PetscViewerASCIIPrintf(viewer,"%% Nonzeros = %d \n",a->nz);
314: PetscViewerASCIIPrintf(viewer,"zzz = zeros(%d,3);\n",a->nz+nofinalvalue);
315: PetscViewerASCIIPrintf(viewer,"zzz = [\n");
317: for (i=0; i<m; i++) {
318: for (j=a->i[i]+shift; j<a->i[i+1]+shift; j++) {
319: #if defined(PETSC_USE_COMPLEX)
320: PetscViewerASCIIPrintf(viewer,"%d %d %18.16e + %18.16ei \n",i+1,a->j[j]+!shift,PetscRealPart(a->a[j]),PetscImaginaryPart(a->a[j]));
321: #else
322: PetscViewerASCIIPrintf(viewer,"%d %d %18.16e\n",i+1,a->j[j]+!shift,a->a[j]);
323: #endif
324: }
325: }
326: if (nofinalvalue) {
327: PetscViewerASCIIPrintf(viewer,"%d %d %18.16e\n",m,A->n,0.0);
328: }
329: PetscViewerASCIIPrintf(viewer,"];\n %s = spconvert(zzz);\n",name);
330: PetscViewerASCIIUseTabs(viewer,PETSC_YES);
331: } else if (format == PETSC_VIEWER_ASCII_FACTOR_INFO) {
332: return(0);
333: } else if (format == PETSC_VIEWER_ASCII_COMMON) {
334: PetscViewerASCIIUseTabs(viewer,PETSC_NO);
335: for (i=0; i<m; i++) {
336: PetscViewerASCIIPrintf(viewer,"row %d:",i);
337: for (j=a->i[i]+shift; j<a->i[i+1]+shift; j++) {
338: #if defined(PETSC_USE_COMPLEX)
339: if (PetscImaginaryPart(a->a[j]) > 0.0 && PetscRealPart(a->a[j]) != 0.0) {
340: PetscViewerASCIIPrintf(viewer," (%d, %g + %g i)",a->j[j]+shift,PetscRealPart(a->a[j]),PetscImaginaryPart(a->a[j]));
341: } else if (PetscImaginaryPart(a->a[j]) < 0.0 && PetscRealPart(a->a[j]) != 0.0) {
342: PetscViewerASCIIPrintf(viewer," (%d, %g - %g i)",a->j[j]+shift,PetscRealPart(a->a[j]),-PetscImaginaryPart(a->a[j]));
343: } else if (PetscRealPart(a->a[j]) != 0.0) {
344: PetscViewerASCIIPrintf(viewer," (%d, %g) ",a->j[j]+shift,PetscRealPart(a->a[j]));
345: }
346: #else
347: if (a->a[j] != 0.0) {PetscViewerASCIIPrintf(viewer," (%d, %g) ",a->j[j]+shift,a->a[j]);}
348: #endif
349: }
350: PetscViewerASCIIPrintf(viewer,"\n");
351: }
352: PetscViewerASCIIUseTabs(viewer,PETSC_YES);
353: } else if (format == PETSC_VIEWER_ASCII_SYMMODU) {
354: int nzd=0,fshift=1,*sptr;
355: PetscViewerASCIIUseTabs(viewer,PETSC_NO);
356: PetscMalloc((m+1)*sizeof(int),&sptr);
357: for (i=0; i<m; i++) {
358: sptr[i] = nzd+1;
359: for (j=a->i[i]+shift; j<a->i[i+1]+shift; j++) {
360: if (a->j[j] >= i) {
361: #if defined(PETSC_USE_COMPLEX)
362: if (PetscImaginaryPart(a->a[j]) != 0.0 || PetscRealPart(a->a[j]) != 0.0) nzd++;
363: #else
364: if (a->a[j] != 0.0) nzd++;
365: #endif
366: }
367: }
368: }
369: sptr[m] = nzd+1;
370: PetscViewerASCIIPrintf(viewer," %d %d\n\n",m,nzd);
371: for (i=0; i<m+1; i+=6) {
372: 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]);}
373: 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]);}
374: else if (i+2<m) {PetscViewerASCIIPrintf(viewer," %d %d %d %d\n",sptr[i],sptr[i+1],sptr[i+2],sptr[i+3]);}
375: else if (i+1<m) {PetscViewerASCIIPrintf(viewer," %d %d %d\n",sptr[i],sptr[i+1],sptr[i+2]);}
376: else if (i<m) {PetscViewerASCIIPrintf(viewer," %d %d\n",sptr[i],sptr[i+1]);}
377: else {PetscViewerASCIIPrintf(viewer," %d\n",sptr[i]);}
378: }
379: PetscViewerASCIIPrintf(viewer,"\n");
380: PetscFree(sptr);
381: for (i=0; i<m; i++) {
382: for (j=a->i[i]+shift; j<a->i[i+1]+shift; j++) {
383: if (a->j[j] >= i) {PetscViewerASCIIPrintf(viewer," %d ",a->j[j]+fshift);}
384: }
385: PetscViewerASCIIPrintf(viewer,"\n");
386: }
387: PetscViewerASCIIPrintf(viewer,"\n");
388: for (i=0; i<m; i++) {
389: for (j=a->i[i]+shift; j<a->i[i+1]+shift; j++) {
390: if (a->j[j] >= i) {
391: #if defined(PETSC_USE_COMPLEX)
392: if (PetscImaginaryPart(a->a[j]) != 0.0 || PetscRealPart(a->a[j]) != 0.0) {
393: PetscViewerASCIIPrintf(viewer," %18.16e %18.16e ",PetscRealPart(a->a[j]),PetscImaginaryPart(a->a[j]));
394: }
395: #else
396: if (a->a[j] != 0.0) {PetscViewerASCIIPrintf(viewer," %18.16e ",a->a[j]);}
397: #endif
398: }
399: }
400: PetscViewerASCIIPrintf(viewer,"\n");
401: }
402: PetscViewerASCIIUseTabs(viewer,PETSC_YES);
403: } else if (format == PETSC_VIEWER_ASCII_DENSE) {
404: int cnt = 0,jcnt;
405: PetscScalar value;
407: PetscViewerASCIIUseTabs(viewer,PETSC_NO);
408: for (i=0; i<m; i++) {
409: jcnt = 0;
410: for (j=0; j<A->n; j++) {
411: if (jcnt < a->i[i+1]-a->i[i] && j == a->j[cnt]) {
412: value = a->a[cnt++];
413: jcnt++;
414: } else {
415: value = 0.0;
416: }
417: #if defined(PETSC_USE_COMPLEX)
418: PetscViewerASCIIPrintf(viewer," %7.5e+%7.5e i ",PetscRealPart(value),PetscImaginaryPart(value));
419: #else
420: PetscViewerASCIIPrintf(viewer," %7.5e ",value);
421: #endif
422: }
423: PetscViewerASCIIPrintf(viewer,"\n");
424: }
425: PetscViewerASCIIUseTabs(viewer,PETSC_YES);
426: } else {
427: PetscViewerASCIIUseTabs(viewer,PETSC_NO);
428: for (i=0; i<m; i++) {
429: PetscViewerASCIIPrintf(viewer,"row %d:",i);
430: for (j=a->i[i]+shift; j<a->i[i+1]+shift; j++) {
431: #if defined(PETSC_USE_COMPLEX)
432: if (PetscImaginaryPart(a->a[j]) > 0.0) {
433: PetscViewerASCIIPrintf(viewer," (%d, %g + %g i)",a->j[j]+shift,PetscRealPart(a->a[j]),PetscImaginaryPart(a->a[j]));
434: } else if (PetscImaginaryPart(a->a[j]) < 0.0) {
435: PetscViewerASCIIPrintf(viewer," (%d, %g - %g i)",a->j[j]+shift,PetscRealPart(a->a[j]),-PetscImaginaryPart(a->a[j]));
436: } else {
437: PetscViewerASCIIPrintf(viewer," (%d, %g) ",a->j[j]+shift,PetscRealPart(a->a[j]));
438: }
439: #else
440: PetscViewerASCIIPrintf(viewer," (%d, %g) ",a->j[j]+shift,a->a[j]);
441: #endif
442: }
443: PetscViewerASCIIPrintf(viewer,"\n");
444: }
445: PetscViewerASCIIUseTabs(viewer,PETSC_YES);
446: }
447: PetscViewerFlush(viewer);
448: return(0);
449: }
453: int MatView_SeqAIJ_Draw_Zoom(PetscDraw draw,void *Aa)
454: {
455: Mat A = (Mat) Aa;
456: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
457: int ierr,i,j,m = A->m,color;
458: PetscReal xl,yl,xr,yr,x_l,x_r,y_l,y_r,maxv = 0.0;
459: PetscViewer viewer;
460: PetscViewerFormat format;
463: PetscObjectQuery((PetscObject)A,"Zoomviewer",(PetscObject*)&viewer);
464: PetscViewerGetFormat(viewer,&format);
466: PetscDrawGetCoordinates(draw,&xl,&yl,&xr,&yr);
467: /* loop over matrix elements drawing boxes */
469: if (format != PETSC_VIEWER_DRAW_CONTOUR) {
470: /* Blue for negative, Cyan for zero and Red for positive */
471: color = PETSC_DRAW_BLUE;
472: for (i=0; i<m; i++) {
473: y_l = m - i - 1.0; y_r = y_l + 1.0;
474: for (j=a->i[i]; j<a->i[i+1]; j++) {
475: x_l = a->j[j] ; x_r = x_l + 1.0;
476: #if defined(PETSC_USE_COMPLEX)
477: if (PetscRealPart(a->a[j]) >= 0.) continue;
478: #else
479: if (a->a[j] >= 0.) continue;
480: #endif
481: PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);
482: }
483: }
484: color = PETSC_DRAW_CYAN;
485: for (i=0; i<m; i++) {
486: y_l = m - i - 1.0; y_r = y_l + 1.0;
487: for (j=a->i[i]; j<a->i[i+1]; j++) {
488: x_l = a->j[j]; x_r = x_l + 1.0;
489: if (a->a[j] != 0.) continue;
490: PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);
491: }
492: }
493: color = PETSC_DRAW_RED;
494: for (i=0; i<m; i++) {
495: y_l = m - i - 1.0; y_r = y_l + 1.0;
496: for (j=a->i[i]; j<a->i[i+1]; j++) {
497: x_l = a->j[j]; x_r = x_l + 1.0;
498: #if defined(PETSC_USE_COMPLEX)
499: if (PetscRealPart(a->a[j]) <= 0.) continue;
500: #else
501: if (a->a[j] <= 0.) continue;
502: #endif
503: PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);
504: }
505: }
506: } else {
507: /* use contour shading to indicate magnitude of values */
508: /* first determine max of all nonzero values */
509: int nz = a->nz,count;
510: PetscDraw popup;
511: PetscReal scale;
513: for (i=0; i<nz; i++) {
514: if (PetscAbsScalar(a->a[i]) > maxv) maxv = PetscAbsScalar(a->a[i]);
515: }
516: scale = (245.0 - PETSC_DRAW_BASIC_COLORS)/maxv;
517: PetscDrawGetPopup(draw,&popup);
518: if (popup) {PetscDrawScalePopup(popup,0.0,maxv);}
519: count = 0;
520: for (i=0; i<m; i++) {
521: y_l = m - i - 1.0; y_r = y_l + 1.0;
522: for (j=a->i[i]; j<a->i[i+1]; j++) {
523: x_l = a->j[j]; x_r = x_l + 1.0;
524: color = PETSC_DRAW_BASIC_COLORS + (int)(scale*PetscAbsScalar(a->a[count]));
525: PetscDrawRectangle(draw,x_l,y_l,x_r,y_r,color,color,color,color);
526: count++;
527: }
528: }
529: }
530: return(0);
531: }
535: int MatView_SeqAIJ_Draw(Mat A,PetscViewer viewer)
536: {
537: int ierr;
538: PetscDraw draw;
539: PetscReal xr,yr,xl,yl,h,w;
540: PetscTruth isnull;
543: PetscViewerDrawGetDraw(viewer,0,&draw);
544: PetscDrawIsNull(draw,&isnull);
545: if (isnull) return(0);
547: PetscObjectCompose((PetscObject)A,"Zoomviewer",(PetscObject)viewer);
548: xr = A->n; yr = A->m; h = yr/10.0; w = xr/10.0;
549: xr += w; yr += h; xl = -w; yl = -h;
550: PetscDrawSetCoordinates(draw,xl,yl,xr,yr);
551: PetscDrawZoom(draw,MatView_SeqAIJ_Draw_Zoom,A);
552: PetscObjectCompose((PetscObject)A,"Zoomviewer",PETSC_NULL);
553: return(0);
554: }
558: int MatView_SeqAIJ(Mat A,PetscViewer viewer)
559: {
560: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
561: int ierr;
562: PetscTruth issocket,isascii,isbinary,isdraw;
565: PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_SOCKET,&issocket);
566: PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_ASCII,&isascii);
567: PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_BINARY,&isbinary);
568: PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_DRAW,&isdraw);
569: if (issocket) {
570: PetscViewerSocketPutSparse_Private(viewer,A->m,A->n,a->nz,a->a,a->i,a->j);
571: } else if (isascii) {
572: MatView_SeqAIJ_ASCII(A,viewer);
573: } else if (isbinary) {
574: MatView_SeqAIJ_Binary(A,viewer);
575: } else if (isdraw) {
576: MatView_SeqAIJ_Draw(A,viewer);
577: } else {
578: SETERRQ1(1,"Viewer type %s not supported by SeqAIJ matrices",((PetscObject)viewer)->type_name);
579: }
580: return(0);
581: }
585: int MatAssemblyEnd_SeqAIJ(Mat A,MatAssemblyType mode)
586: {
587: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
588: int fshift = 0,i,j,*ai = a->i,*aj = a->j,*imax = a->imax,ierr;
589: int m = A->m,*ip,N,*ailen = a->ilen,rmax = 0;
590: PetscScalar *aa = a->a,*ap;
593: if (mode == MAT_FLUSH_ASSEMBLY) return(0);
595: if (m) rmax = ailen[0]; /* determine row with most nonzeros */
596: for (i=1; i<m; i++) {
597: /* move each row back by the amount of empty slots (fshift) before it*/
598: fshift += imax[i-1] - ailen[i-1];
599: rmax = PetscMax(rmax,ailen[i]);
600: if (fshift) {
601: ip = aj + ai[i] ;
602: ap = aa + ai[i] ;
603: N = ailen[i];
604: for (j=0; j<N; j++) {
605: ip[j-fshift] = ip[j];
606: ap[j-fshift] = ap[j];
607: }
608: }
609: ai[i] = ai[i-1] + ailen[i-1];
610: }
611: if (m) {
612: fshift += imax[m-1] - ailen[m-1];
613: ai[m] = ai[m-1] + ailen[m-1];
614: }
615: /* reset ilen and imax for each row */
616: for (i=0; i<m; i++) {
617: ailen[i] = imax[i] = ai[i+1] - ai[i];
618: }
619: a->nz = ai[m];
621: /* diagonals may have moved, so kill the diagonal pointers */
622: if (fshift && a->diag) {
623: PetscFree(a->diag);
624: PetscLogObjectMemory(A,-(m+1)*sizeof(int));
625: a->diag = 0;
626: }
627: PetscLogInfo(A,"MatAssemblyEnd_SeqAIJ:Matrix size: %d X %d; storage space: %d unneeded,%d used\n",m,A->n,fshift,a->nz);
628: PetscLogInfo(A,"MatAssemblyEnd_SeqAIJ:Number of mallocs during MatSetValues() is %d\n",a->reallocs);
629: PetscLogInfo(A,"MatAssemblyEnd_SeqAIJ:Most nonzeros in any row is %d\n",rmax);
630: a->reallocs = 0;
631: A->info.nz_unneeded = (double)fshift;
632: a->rmax = rmax;
634: /* check out for identical nodes. If found, use inode functions */
635: Mat_AIJ_CheckInode(A,(PetscTruth)(!fshift));
637: return(0);
638: }
642: int MatZeroEntries_SeqAIJ(Mat A)
643: {
644: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
645: int ierr;
648: PetscMemzero(a->a,(a->i[A->m])*sizeof(PetscScalar));
649: return(0);
650: }
654: int MatDestroy_SeqAIJ(Mat A)
655: {
656: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
657: int ierr;
660: #if defined(PETSC_USE_LOG)
661: PetscLogObjectState((PetscObject)A,"Rows=%d, Cols=%d, NZ=%d",A->m,A->n,a->nz);
662: #endif
663: if (a->freedata) {
664: PetscFree(a->a);
665: if (!a->singlemalloc) {
666: PetscFree(a->i);
667: PetscFree(a->j);
668: }
669: }
670: if (a->row) {
671: ISDestroy(a->row);
672: }
673: if (a->col) {
674: ISDestroy(a->col);
675: }
676: if (a->diag) {PetscFree(a->diag);}
677: if (a->ilen) {PetscFree(a->ilen);}
678: if (a->imax) {PetscFree(a->imax);}
679: if (a->idiag) {PetscFree(a->idiag);}
680: if (a->solve_work) {PetscFree(a->solve_work);}
681: if (a->inode.size) {PetscFree(a->inode.size);}
682: if (a->icol) {ISDestroy(a->icol);}
683: if (a->saved_values) {PetscFree(a->saved_values);}
684: if (a->coloring) {ISColoringDestroy(a->coloring);}
685: if (a->xtoy) {PetscFree(a->xtoy);}
686:
687: PetscFree(a);
688: return(0);
689: }
693: int MatCompress_SeqAIJ(Mat A)
694: {
696: return(0);
697: }
701: int MatSetOption_SeqAIJ(Mat A,MatOption op)
702: {
703: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
706: switch (op) {
707: case MAT_ROW_ORIENTED:
708: a->roworiented = PETSC_TRUE;
709: break;
710: case MAT_KEEP_ZEROED_ROWS:
711: a->keepzeroedrows = PETSC_TRUE;
712: break;
713: case MAT_COLUMN_ORIENTED:
714: a->roworiented = PETSC_FALSE;
715: break;
716: case MAT_COLUMNS_SORTED:
717: a->sorted = PETSC_TRUE;
718: break;
719: case MAT_COLUMNS_UNSORTED:
720: a->sorted = PETSC_FALSE;
721: break;
722: case MAT_NO_NEW_NONZERO_LOCATIONS:
723: a->nonew = 1;
724: break;
725: case MAT_NEW_NONZERO_LOCATION_ERR:
726: a->nonew = -1;
727: break;
728: case MAT_NEW_NONZERO_ALLOCATION_ERR:
729: a->nonew = -2;
730: break;
731: case MAT_YES_NEW_NONZERO_LOCATIONS:
732: a->nonew = 0;
733: break;
734: case MAT_IGNORE_ZERO_ENTRIES:
735: a->ignorezeroentries = PETSC_TRUE;
736: break;
737: case MAT_USE_INODES:
738: a->inode.use = PETSC_TRUE;
739: break;
740: case MAT_DO_NOT_USE_INODES:
741: a->inode.use = PETSC_FALSE;
742: break;
743: case MAT_ROWS_SORTED:
744: case MAT_ROWS_UNSORTED:
745: case MAT_YES_NEW_DIAGONALS:
746: case MAT_IGNORE_OFF_PROC_ENTRIES:
747: case MAT_USE_HASH_TABLE:
748: PetscLogInfo(A,"MatSetOption_SeqAIJ:Option ignored\n");
749: break;
750: case MAT_NO_NEW_DIAGONALS:
751: SETERRQ(PETSC_ERR_SUP,"MAT_NO_NEW_DIAGONALS");
752: case MAT_INODE_LIMIT_1:
753: a->inode.limit = 1;
754: break;
755: case MAT_INODE_LIMIT_2:
756: a->inode.limit = 2;
757: break;
758: case MAT_INODE_LIMIT_3:
759: a->inode.limit = 3;
760: break;
761: case MAT_INODE_LIMIT_4:
762: a->inode.limit = 4;
763: break;
764: case MAT_INODE_LIMIT_5:
765: a->inode.limit = 5;
766: break;
767: case MAT_SYMMETRIC:
768: case MAT_STRUCTURALLY_SYMMETRIC:
769: case MAT_NOT_SYMMETRIC:
770: case MAT_NOT_STRUCTURALLY_SYMMETRIC:
771: case MAT_HERMITIAN:
772: case MAT_NOT_HERMITIAN:
773: case MAT_SYMMETRY_ETERNAL:
774: case MAT_NOT_SYMMETRY_ETERNAL:
775: break;
776: default:
777: SETERRQ(PETSC_ERR_SUP,"unknown option");
778: }
779: return(0);
780: }
784: int MatGetDiagonal_SeqAIJ(Mat A,Vec v)
785: {
786: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
787: int i,j,n,ierr;
788: PetscScalar *x,zero = 0.0;
791: VecSet(&zero,v);
792: VecGetArray(v,&x);
793: VecGetLocalSize(v,&n);
794: if (n != A->m) SETERRQ(PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector");
795: for (i=0; i<A->m; i++) {
796: for (j=a->i[i]; j<a->i[i+1]; j++) {
797: if (a->j[j] == i) {
798: x[i] = a->a[j];
799: break;
800: }
801: }
802: }
803: VecRestoreArray(v,&x);
804: return(0);
805: }
810: int MatMultTransposeAdd_SeqAIJ(Mat A,Vec xx,Vec zz,Vec yy)
811: {
812: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
813: PetscScalar *x,*y;
814: int ierr,m = A->m;
815: #if !defined(PETSC_USE_FORTRAN_KERNEL_MULTTRANSPOSEAIJ)
816: PetscScalar *v,alpha;
817: int n,i,*idx;
818: #endif
821: if (zz != yy) {VecCopy(zz,yy);}
822: VecGetArray(xx,&x);
823: VecGetArray(yy,&y);
825: #if defined(PETSC_USE_FORTRAN_KERNEL_MULTTRANSPOSEAIJ)
826: fortranmulttransposeaddaij_(&m,x,a->i,a->j,a->a,y);
827: #else
828: for (i=0; i<m; i++) {
829: idx = a->j + a->i[i] ;
830: v = a->a + a->i[i] ;
831: n = a->i[i+1] - a->i[i];
832: alpha = x[i];
833: while (n-->0) {y[*idx++] += alpha * *v++;}
834: }
835: #endif
836: PetscLogFlops(2*a->nz);
837: VecRestoreArray(xx,&x);
838: VecRestoreArray(yy,&y);
839: return(0);
840: }
844: int MatMultTranspose_SeqAIJ(Mat A,Vec xx,Vec yy)
845: {
846: PetscScalar zero = 0.0;
847: int ierr;
850: VecSet(&zero,yy);
851: MatMultTransposeAdd_SeqAIJ(A,xx,yy,yy);
852: return(0);
853: }
858: int MatMult_SeqAIJ(Mat A,Vec xx,Vec yy)
859: {
860: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
861: PetscScalar *x,*y,*v;
862: int ierr,m = A->m,*idx,*ii;
863: #if !defined(PETSC_USE_FORTRAN_KERNEL_MULTAIJ)
864: int n,i,jrow,j;
865: PetscScalar sum;
866: #endif
868: #if defined(PETSC_HAVE_PRAGMA_DISJOINT)
869: #pragma disjoint(*x,*y,*v)
870: #endif
873: VecGetArray(xx,&x);
874: VecGetArray(yy,&y);
875: idx = a->j;
876: v = a->a;
877: ii = a->i;
878: #if defined(PETSC_USE_FORTRAN_KERNEL_MULTAIJ)
879: fortranmultaij_(&m,x,ii,idx,v,y);
880: #else
881: for (i=0; i<m; i++) {
882: jrow = ii[i];
883: n = ii[i+1] - jrow;
884: sum = 0.0;
885: for (j=0; j<n; j++) {
886: sum += v[jrow]*x[idx[jrow]]; jrow++;
887: }
888: y[i] = sum;
889: }
890: #endif
891: PetscLogFlops(2*a->nz - m);
892: VecRestoreArray(xx,&x);
893: VecRestoreArray(yy,&y);
894: return(0);
895: }
899: int MatMultAdd_SeqAIJ(Mat A,Vec xx,Vec yy,Vec zz)
900: {
901: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
902: PetscScalar *x,*y,*z,*v;
903: int ierr,m = A->m,*idx,*ii;
904: #if !defined(PETSC_USE_FORTRAN_KERNEL_MULTADDAIJ)
905: int n,i,jrow,j;
906: PetscScalar sum;
907: #endif
910: VecGetArray(xx,&x);
911: VecGetArray(yy,&y);
912: if (zz != yy) {
913: VecGetArray(zz,&z);
914: } else {
915: z = y;
916: }
917:
918: idx = a->j;
919: v = a->a;
920: ii = a->i;
921: #if defined(PETSC_USE_FORTRAN_KERNEL_MULTADDAIJ)
922: fortranmultaddaij_(&m,x,ii,idx,v,y,z);
923: #else
924: for (i=0; i<m; i++) {
925: jrow = ii[i];
926: n = ii[i+1] - jrow;
927: sum = y[i];
928: for (j=0; j<n; j++) {
929: sum += v[jrow]*x[idx[jrow]]; jrow++;
930: }
931: z[i] = sum;
932: }
933: #endif
934: PetscLogFlops(2*a->nz);
935: VecRestoreArray(xx,&x);
936: VecRestoreArray(yy,&y);
937: if (zz != yy) {
938: VecRestoreArray(zz,&z);
939: }
940: return(0);
941: }
943: /*
944: Adds diagonal pointers to sparse matrix structure.
945: */
948: int MatMarkDiagonal_SeqAIJ(Mat A)
949: {
950: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
951: int i,j,*diag,m = A->m,ierr;
954: if (a->diag) return(0);
956: PetscMalloc((m+1)*sizeof(int),&diag);
957: PetscLogObjectMemory(A,(m+1)*sizeof(int));
958: for (i=0; i<A->m; i++) {
959: diag[i] = a->i[i+1];
960: for (j=a->i[i]; j<a->i[i+1]; j++) {
961: if (a->j[j] == i) {
962: diag[i] = j;
963: break;
964: }
965: }
966: }
967: a->diag = diag;
968: return(0);
969: }
971: /*
972: Checks for missing diagonals
973: */
976: int MatMissingDiagonal_SeqAIJ(Mat A)
977: {
978: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
979: int *diag,*jj = a->j,i,ierr;
982: MatMarkDiagonal_SeqAIJ(A);
983: diag = a->diag;
984: for (i=0; i<A->m; i++) {
985: if (jj[diag[i]] != i) {
986: SETERRQ1(1,"Matrix is missing diagonal number %d",i);
987: }
988: }
989: return(0);
990: }
994: int MatRelax_SeqAIJ(Mat A,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,int its,int lits,Vec xx)
995: {
996: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
997: PetscScalar *x,d,*xs,sum,*t,scale,*idiag=0,*mdiag;
998: const PetscScalar *v = a->a, *b, *bs,*xb, *ts;
999: int ierr,n = A->n,m = A->m,i;
1000: const int *idx,*diag;
1003: its = its*lits;
1004: if (its <= 0) SETERRQ2(PETSC_ERR_ARG_WRONG,"Relaxation requires global its %d and local its %d both positive",its,lits);
1006: if (!a->diag) {MatMarkDiagonal_SeqAIJ(A);}
1007: diag = a->diag;
1008: if (!a->idiag) {
1009: PetscMalloc(3*m*sizeof(PetscScalar),&a->idiag);
1010: a->ssor = a->idiag + m;
1011: mdiag = a->ssor + m;
1013: v = a->a;
1015: /* this is wrong when fshift omega changes each iteration */
1016: if (omega == 1.0 && fshift == 0.0) {
1017: for (i=0; i<m; i++) {
1018: mdiag[i] = v[diag[i]];
1019: a->idiag[i] = 1.0/v[diag[i]];
1020: }
1021: PetscLogFlops(m);
1022: } else {
1023: for (i=0; i<m; i++) {
1024: mdiag[i] = v[diag[i]];
1025: a->idiag[i] = omega/(fshift + v[diag[i]]);
1026: }
1027: PetscLogFlops(2*m);
1028: }
1029: }
1030: t = a->ssor;
1031: idiag = a->idiag;
1032: mdiag = a->idiag + 2*m;
1034: VecGetArray(xx,&x);
1035: if (xx != bb) {
1036: VecGetArray(bb,(PetscScalar**)&b);
1037: } else {
1038: b = x;
1039: }
1041: /* We count flops by assuming the upper triangular and lower triangular parts have the same number of nonzeros */
1042: xs = x;
1043: if (flag == SOR_APPLY_UPPER) {
1044: /* apply (U + D/omega) to the vector */
1045: bs = b;
1046: for (i=0; i<m; i++) {
1047: d = fshift + a->a[diag[i]];
1048: n = a->i[i+1] - diag[i] - 1;
1049: idx = a->j + diag[i] + 1;
1050: v = a->a + diag[i] + 1;
1051: sum = b[i]*d/omega;
1052: SPARSEDENSEDOT(sum,bs,v,idx,n);
1053: x[i] = sum;
1054: }
1055: VecRestoreArray(xx,&x);
1056: if (bb != xx) {VecRestoreArray(bb,(PetscScalar**)&b);}
1057: PetscLogFlops(a->nz);
1058: return(0);
1059: }
1062: /* Let A = L + U + D; where L is lower trianglar,
1063: U is upper triangular, E is diagonal; This routine applies
1065: (L + E)^{-1} A (U + E)^{-1}
1067: to a vector efficiently using Eisenstat's trick. This is for
1068: the case of SSOR preconditioner, so E is D/omega where omega
1069: is the relaxation factor.
1070: */
1072: if (flag == SOR_APPLY_LOWER) {
1073: SETERRQ(PETSC_ERR_SUP,"SOR_APPLY_LOWER is not implemented");
1074: } else if (flag & SOR_EISENSTAT) {
1075: /* Let A = L + U + D; where L is lower trianglar,
1076: U is upper triangular, E is diagonal; This routine applies
1078: (L + E)^{-1} A (U + E)^{-1}
1080: to a vector efficiently using Eisenstat's trick. This is for
1081: the case of SSOR preconditioner, so E is D/omega where omega
1082: is the relaxation factor.
1083: */
1084: scale = (2.0/omega) - 1.0;
1086: /* x = (E + U)^{-1} b */
1087: for (i=m-1; i>=0; i--) {
1088: n = a->i[i+1] - diag[i] - 1;
1089: idx = a->j + diag[i] + 1;
1090: v = a->a + diag[i] + 1;
1091: sum = b[i];
1092: SPARSEDENSEMDOT(sum,xs,v,idx,n);
1093: x[i] = sum*idiag[i];
1094: }
1096: /* t = b - (2*E - D)x */
1097: v = a->a;
1098: for (i=0; i<m; i++) { t[i] = b[i] - scale*(v[*diag++])*x[i]; }
1100: /* t = (E + L)^{-1}t */
1101: ts = t;
1102: diag = a->diag;
1103: for (i=0; i<m; i++) {
1104: n = diag[i] - a->i[i];
1105: idx = a->j + a->i[i];
1106: v = a->a + a->i[i];
1107: sum = t[i];
1108: SPARSEDENSEMDOT(sum,ts,v,idx,n);
1109: t[i] = sum*idiag[i];
1110: /* x = x + t */
1111: x[i] += t[i];
1112: }
1114: PetscLogFlops(6*m-1 + 2*a->nz);
1115: VecRestoreArray(xx,&x);
1116: if (bb != xx) {VecRestoreArray(bb,(PetscScalar**)&b);}
1117: return(0);
1118: }
1119: if (flag & SOR_ZERO_INITIAL_GUESS) {
1120: if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP){
1121: #if defined(PETSC_USE_FORTRAN_KERNEL_RELAXAIJ)
1122: fortranrelaxaijforwardzero_(&m,&omega,x,a->i,a->j,(int *)diag,idiag,a->a,(void*)b);
1123: #else
1124: for (i=0; i<m; i++) {
1125: n = diag[i] - a->i[i];
1126: idx = a->j + a->i[i];
1127: v = a->a + a->i[i];
1128: sum = b[i];
1129: SPARSEDENSEMDOT(sum,xs,v,idx,n);
1130: x[i] = sum*idiag[i];
1131: }
1132: #endif
1133: xb = x;
1134: PetscLogFlops(a->nz);
1135: } else xb = b;
1136: if ((flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) &&
1137: (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP)) {
1138: for (i=0; i<m; i++) {
1139: x[i] *= mdiag[i];
1140: }
1141: PetscLogFlops(m);
1142: }
1143: if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP){
1144: #if defined(PETSC_USE_FORTRAN_KERNEL_RELAXAIJ)
1145: fortranrelaxaijbackwardzero_(&m,&omega,x,a->i,a->j,(int*)diag,idiag,a->a,(void*)xb);
1146: #else
1147: for (i=m-1; i>=0; i--) {
1148: n = a->i[i+1] - diag[i] - 1;
1149: idx = a->j + diag[i] + 1;
1150: v = a->a + diag[i] + 1;
1151: sum = xb[i];
1152: SPARSEDENSEMDOT(sum,xs,v,idx,n);
1153: x[i] = sum*idiag[i];
1154: }
1155: #endif
1156: PetscLogFlops(a->nz);
1157: }
1158: its--;
1159: }
1160: while (its--) {
1161: if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP){
1162: #if defined(PETSC_USE_FORTRAN_KERNEL_RELAXAIJ)
1163: fortranrelaxaijforward_(&m,&omega,x,a->i,a->j,(int*)diag,a->a,(void*)b);
1164: #else
1165: for (i=0; i<m; i++) {
1166: d = fshift + a->a[diag[i]];
1167: n = a->i[i+1] - a->i[i];
1168: idx = a->j + a->i[i];
1169: v = a->a + a->i[i];
1170: sum = b[i];
1171: SPARSEDENSEMDOT(sum,xs,v,idx,n);
1172: x[i] = (1. - omega)*x[i] + (sum + mdiag[i]*x[i])*idiag[i];
1173: }
1174: #endif
1175: PetscLogFlops(a->nz);
1176: }
1177: if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP){
1178: #if defined(PETSC_USE_FORTRAN_KERNEL_RELAXAIJ)
1179: fortranrelaxaijbackward_(&m,&omega,x,a->i,a->j,(int*)diag,a->a,(void*)b);
1180: #else
1181: for (i=m-1; i>=0; i--) {
1182: d = fshift + a->a[diag[i]];
1183: n = a->i[i+1] - a->i[i];
1184: idx = a->j + a->i[i];
1185: v = a->a + a->i[i];
1186: sum = b[i];
1187: SPARSEDENSEMDOT(sum,xs,v,idx,n);
1188: x[i] = (1. - omega)*x[i] + (sum + mdiag[i]*x[i])*idiag[i];
1189: }
1190: #endif
1191: PetscLogFlops(a->nz);
1192: }
1193: }
1194: VecRestoreArray(xx,&x);
1195: if (bb != xx) {VecRestoreArray(bb,(PetscScalar**)&b);}
1196: return(0);
1197: }
1201: int MatGetInfo_SeqAIJ(Mat A,MatInfoType flag,MatInfo *info)
1202: {
1203: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1206: info->rows_global = (double)A->m;
1207: info->columns_global = (double)A->n;
1208: info->rows_local = (double)A->m;
1209: info->columns_local = (double)A->n;
1210: info->block_size = 1.0;
1211: info->nz_allocated = (double)a->maxnz;
1212: info->nz_used = (double)a->nz;
1213: info->nz_unneeded = (double)(a->maxnz - a->nz);
1214: info->assemblies = (double)A->num_ass;
1215: info->mallocs = (double)a->reallocs;
1216: info->memory = A->mem;
1217: if (A->factor) {
1218: info->fill_ratio_given = A->info.fill_ratio_given;
1219: info->fill_ratio_needed = A->info.fill_ratio_needed;
1220: info->factor_mallocs = A->info.factor_mallocs;
1221: } else {
1222: info->fill_ratio_given = 0;
1223: info->fill_ratio_needed = 0;
1224: info->factor_mallocs = 0;
1225: }
1226: return(0);
1227: }
1231: int MatZeroRows_SeqAIJ(Mat A,IS is,const PetscScalar *diag)
1232: {
1233: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1234: int i,ierr,N,*rows,m = A->m - 1;
1237: ISGetLocalSize(is,&N);
1238: ISGetIndices(is,&rows);
1239: if (a->keepzeroedrows) {
1240: for (i=0; i<N; i++) {
1241: if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"row %d out of range", rows[i]);
1242: PetscMemzero(&a->a[a->i[rows[i]]],a->ilen[rows[i]]*sizeof(PetscScalar));
1243: }
1244: if (diag) {
1245: MatMissingDiagonal_SeqAIJ(A);
1246: MatMarkDiagonal_SeqAIJ(A);
1247: for (i=0; i<N; i++) {
1248: a->a[a->diag[rows[i]]] = *diag;
1249: }
1250: }
1251: } else {
1252: if (diag) {
1253: for (i=0; i<N; i++) {
1254: if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"row %d out of range", rows[i]);
1255: if (a->ilen[rows[i]] > 0) {
1256: a->ilen[rows[i]] = 1;
1257: a->a[a->i[rows[i]]] = *diag;
1258: a->j[a->i[rows[i]]] = rows[i];
1259: } else { /* in case row was completely empty */
1260: MatSetValues_SeqAIJ(A,1,&rows[i],1,&rows[i],diag,INSERT_VALUES);
1261: }
1262: }
1263: } else {
1264: for (i=0; i<N; i++) {
1265: if (rows[i] < 0 || rows[i] > m) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"row %d out of range", rows[i]);
1266: a->ilen[rows[i]] = 0;
1267: }
1268: }
1269: }
1270: ISRestoreIndices(is,&rows);
1271: MatAssemblyEnd_SeqAIJ(A,MAT_FINAL_ASSEMBLY);
1272: return(0);
1273: }
1277: int MatGetRow_SeqAIJ(Mat A,int row,int *nz,int **idx,PetscScalar **v)
1278: {
1279: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1280: int *itmp;
1283: if (row < 0 || row >= A->m) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"Row %d out of range",row);
1285: *nz = a->i[row+1] - a->i[row];
1286: if (v) *v = a->a + a->i[row];
1287: if (idx) {
1288: itmp = a->j + a->i[row];
1289: if (*nz) {
1290: *idx = itmp;
1291: }
1292: else *idx = 0;
1293: }
1294: return(0);
1295: }
1297: /* remove this function? */
1300: int MatRestoreRow_SeqAIJ(Mat A,int row,int *nz,int **idx,PetscScalar **v)
1301: {
1302: /* Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1306: /* if (idx) {if (*idx && a->indexshift) {PetscFree(*idx);}} */
1307: return(0);
1308: }
1312: int MatNorm_SeqAIJ(Mat A,NormType type,PetscReal *nrm)
1313: {
1314: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1315: PetscScalar *v = a->a;
1316: PetscReal sum = 0.0;
1317: int i,j,ierr;
1320: if (type == NORM_FROBENIUS) {
1321: for (i=0; i<a->nz; i++) {
1322: #if defined(PETSC_USE_COMPLEX)
1323: sum += PetscRealPart(PetscConj(*v)*(*v)); v++;
1324: #else
1325: sum += (*v)*(*v); v++;
1326: #endif
1327: }
1328: *nrm = sqrt(sum);
1329: } else if (type == NORM_1) {
1330: PetscReal *tmp;
1331: int *jj = a->j;
1332: PetscMalloc((A->n+1)*sizeof(PetscReal),&tmp);
1333: PetscMemzero(tmp,A->n*sizeof(PetscReal));
1334: *nrm = 0.0;
1335: for (j=0; j<a->nz; j++) {
1336: tmp[*jj++] += PetscAbsScalar(*v); v++;
1337: }
1338: for (j=0; j<A->n; j++) {
1339: if (tmp[j] > *nrm) *nrm = tmp[j];
1340: }
1341: PetscFree(tmp);
1342: } else if (type == NORM_INFINITY) {
1343: *nrm = 0.0;
1344: for (j=0; j<A->m; j++) {
1345: v = a->a + a->i[j];
1346: sum = 0.0;
1347: for (i=0; i<a->i[j+1]-a->i[j]; i++) {
1348: sum += PetscAbsScalar(*v); v++;
1349: }
1350: if (sum > *nrm) *nrm = sum;
1351: }
1352: } else {
1353: SETERRQ(PETSC_ERR_SUP,"No support for two norm");
1354: }
1355: return(0);
1356: }
1360: int MatTranspose_SeqAIJ(Mat A,Mat *B)
1361: {
1362: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1363: Mat C;
1364: int i,ierr,*aj = a->j,*ai = a->i,m = A->m,len,*col;
1365: PetscScalar *array = a->a;
1368: if (!B && m != A->n) SETERRQ(PETSC_ERR_ARG_SIZ,"Square matrix only for in-place");
1369: PetscMalloc((1+A->n)*sizeof(int),&col);
1370: PetscMemzero(col,(1+A->n)*sizeof(int));
1371:
1372: for (i=0; i<ai[m]; i++) col[aj[i]] += 1;
1373: MatCreate(A->comm,A->n,m,A->n,m,&C);
1374: MatSetType(C,A->type_name);
1375: MatSeqAIJSetPreallocation(C,0,col);
1376: PetscFree(col);
1377: for (i=0; i<m; i++) {
1378: len = ai[i+1]-ai[i];
1379: MatSetValues(C,len,aj,1,&i,array,INSERT_VALUES);
1380: array += len;
1381: aj += len;
1382: }
1384: MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);
1385: MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);
1387: if (B) {
1388: *B = C;
1389: } else {
1390: MatHeaderCopy(A,C);
1391: }
1392: return(0);
1393: }
1395: EXTERN_C_BEGIN
1398: int MatIsTranspose_SeqAIJ(Mat A,Mat B,PetscTruth *f)
1399: {
1400: Mat_SeqAIJ *aij = (Mat_SeqAIJ *) A->data,*bij = (Mat_SeqAIJ*) A->data;
1401: int *adx,*bdx,*aii,*bii,*aptr,*bptr; PetscScalar *va,*vb;
1402: int ma,na,mb,nb, i,ierr;
1405: bij = (Mat_SeqAIJ *) B->data;
1406:
1407: MatGetSize(A,&ma,&na);
1408: MatGetSize(B,&mb,&nb);
1409: if (ma!=nb || na!=mb)
1410: SETERRQ(1,"Incompatible A/B sizes for symmetry test");
1411: aii = aij->i; bii = bij->i;
1412: adx = aij->j; bdx = bij->j;
1413: va = aij->a; vb = bij->a;
1414: PetscMalloc(ma*sizeof(int),&aptr);
1415: PetscMalloc(mb*sizeof(int),&bptr);
1416: for (i=0; i<ma; i++) aptr[i] = aii[i];
1417: for (i=0; i<mb; i++) bptr[i] = bii[i];
1419: *f = PETSC_TRUE;
1420: for (i=0; i<ma; i++) {
1421: /*printf("row %d spans %d--%d; we start @ %d\n",
1422: i,idx[ii[i]],idx[ii[i+1]-1],idx[aptr[i]]);*/
1423: while (aptr[i]<aii[i+1]) {
1424: int idc,idr; PetscScalar vc,vr;
1425: /* column/row index/value */
1426: idc = adx[aptr[i]]; idr = bdx[bptr[idc]];
1427: vc = va[aptr[i]]; vr = vb[bptr[idc]];
1428: /*printf("comparing %d: (%d,%d)=%e to (%d,%d)=%e\n",
1429: aptr[i],i,idc,vc,idc,idr,vr);*/
1430: if (i!=idr || vc!=vr) {
1431: *f = PETSC_FALSE; goto done;
1432: } else {
1433: aptr[i]++; if (B || i!=idc) bptr[idc]++;
1434: }
1435: }
1436: }
1437: done:
1438: PetscFree(aptr);
1439: if (B) {
1440: PetscFree(bptr);
1441: }
1443: return(0);
1444: }
1445: EXTERN_C_END
1449: int MatIsSymmetric_SeqAIJ(Mat A,PetscTruth *f)
1450: {
1453: MatIsTranspose_SeqAIJ(A,A,f);
1454: return(0);
1455: }
1459: int MatDiagonalScale_SeqAIJ(Mat A,Vec ll,Vec rr)
1460: {
1461: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1462: PetscScalar *l,*r,x,*v;
1463: int ierr,i,j,m = A->m,n = A->n,M,nz = a->nz,*jj;
1466: if (ll) {
1467: /* The local size is used so that VecMPI can be passed to this routine
1468: by MatDiagonalScale_MPIAIJ */
1469: VecGetLocalSize(ll,&m);
1470: if (m != A->m) SETERRQ(PETSC_ERR_ARG_SIZ,"Left scaling vector wrong length");
1471: VecGetArray(ll,&l);
1472: v = a->a;
1473: for (i=0; i<m; i++) {
1474: x = l[i];
1475: M = a->i[i+1] - a->i[i];
1476: for (j=0; j<M; j++) { (*v++) *= x;}
1477: }
1478: VecRestoreArray(ll,&l);
1479: PetscLogFlops(nz);
1480: }
1481: if (rr) {
1482: VecGetLocalSize(rr,&n);
1483: if (n != A->n) SETERRQ(PETSC_ERR_ARG_SIZ,"Right scaling vector wrong length");
1484: VecGetArray(rr,&r);
1485: v = a->a; jj = a->j;
1486: for (i=0; i<nz; i++) {
1487: (*v++) *= r[*jj++];
1488: }
1489: VecRestoreArray(rr,&r);
1490: PetscLogFlops(nz);
1491: }
1492: return(0);
1493: }
1497: int MatGetSubMatrix_SeqAIJ(Mat A,IS isrow,IS iscol,int csize,MatReuse scall,Mat *B)
1498: {
1499: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data,*c;
1500: int *smap,i,k,kstart,kend,ierr,oldcols = A->n,*lens;
1501: int row,mat_i,*mat_j,tcol,first,step,*mat_ilen,sum,lensi;
1502: int *irow,*icol,nrows,ncols;
1503: int *starts,*j_new,*i_new,*aj = a->j,*ai = a->i,ii,*ailen = a->ilen;
1504: PetscScalar *a_new,*mat_a;
1505: Mat C;
1506: PetscTruth stride;
1509: ISSorted(isrow,(PetscTruth*)&i);
1510: if (!i) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"ISrow is not sorted");
1511: ISSorted(iscol,(PetscTruth*)&i);
1512: if (!i) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"IScol is not sorted");
1514: ISGetIndices(isrow,&irow);
1515: ISGetLocalSize(isrow,&nrows);
1516: ISGetLocalSize(iscol,&ncols);
1518: ISStrideGetInfo(iscol,&first,&step);
1519: ISStride(iscol,&stride);
1520: if (stride && step == 1) {
1521: /* special case of contiguous rows */
1522: PetscMalloc((2*nrows+1)*sizeof(int),&lens);
1523: starts = lens + nrows;
1524: /* loop over new rows determining lens and starting points */
1525: for (i=0; i<nrows; i++) {
1526: kstart = ai[irow[i]];
1527: kend = kstart + ailen[irow[i]];
1528: for (k=kstart; k<kend; k++) {
1529: if (aj[k] >= first) {
1530: starts[i] = k;
1531: break;
1532: }
1533: }
1534: sum = 0;
1535: while (k < kend) {
1536: if (aj[k++] >= first+ncols) break;
1537: sum++;
1538: }
1539: lens[i] = sum;
1540: }
1541: /* create submatrix */
1542: if (scall == MAT_REUSE_MATRIX) {
1543: int n_cols,n_rows;
1544: MatGetSize(*B,&n_rows,&n_cols);
1545: if (n_rows != nrows || n_cols != ncols) SETERRQ(PETSC_ERR_ARG_SIZ,"Reused submatrix wrong size");
1546: MatZeroEntries(*B);
1547: C = *B;
1548: } else {
1549: MatCreate(A->comm,nrows,ncols,PETSC_DETERMINE,PETSC_DETERMINE,&C);
1550: MatSetType(C,A->type_name);
1551: MatSeqAIJSetPreallocation(C,0,lens);
1552: }
1553: c = (Mat_SeqAIJ*)C->data;
1555: /* loop over rows inserting into submatrix */
1556: a_new = c->a;
1557: j_new = c->j;
1558: i_new = c->i;
1560: for (i=0; i<nrows; i++) {
1561: ii = starts[i];
1562: lensi = lens[i];
1563: for (k=0; k<lensi; k++) {
1564: *j_new++ = aj[ii+k] - first;
1565: }
1566: PetscMemcpy(a_new,a->a + starts[i],lensi*sizeof(PetscScalar));
1567: a_new += lensi;
1568: i_new[i+1] = i_new[i] + lensi;
1569: c->ilen[i] = lensi;
1570: }
1571: PetscFree(lens);
1572: } else {
1573: ISGetIndices(iscol,&icol);
1574: PetscMalloc((1+oldcols)*sizeof(int),&smap);
1575:
1576: PetscMalloc((1+nrows)*sizeof(int),&lens);
1577: PetscMemzero(smap,oldcols*sizeof(int));
1578: for (i=0; i<ncols; i++) smap[icol[i]] = i+1;
1579: /* determine lens of each row */
1580: for (i=0; i<nrows; i++) {
1581: kstart = ai[irow[i]];
1582: kend = kstart + a->ilen[irow[i]];
1583: lens[i] = 0;
1584: for (k=kstart; k<kend; k++) {
1585: if (smap[aj[k]]) {
1586: lens[i]++;
1587: }
1588: }
1589: }
1590: /* Create and fill new matrix */
1591: if (scall == MAT_REUSE_MATRIX) {
1592: PetscTruth equal;
1594: c = (Mat_SeqAIJ *)((*B)->data);
1595: if ((*B)->m != nrows || (*B)->n != ncols) SETERRQ(PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. wrong size");
1596: PetscMemcmp(c->ilen,lens,(*B)->m*sizeof(int),&equal);
1597: if (!equal) {
1598: SETERRQ(PETSC_ERR_ARG_SIZ,"Cannot reuse matrix. wrong no of nonzeros");
1599: }
1600: PetscMemzero(c->ilen,(*B)->m*sizeof(int));
1601: C = *B;
1602: } else {
1603: MatCreate(A->comm,nrows,ncols,PETSC_DETERMINE,PETSC_DETERMINE,&C);
1604: MatSetType(C,A->type_name);
1605: MatSeqAIJSetPreallocation(C,0,lens);
1606: }
1607: c = (Mat_SeqAIJ *)(C->data);
1608: for (i=0; i<nrows; i++) {
1609: row = irow[i];
1610: kstart = ai[row];
1611: kend = kstart + a->ilen[row];
1612: mat_i = c->i[i];
1613: mat_j = c->j + mat_i;
1614: mat_a = c->a + mat_i;
1615: mat_ilen = c->ilen + i;
1616: for (k=kstart; k<kend; k++) {
1617: if ((tcol=smap[a->j[k]])) {
1618: *mat_j++ = tcol - 1;
1619: *mat_a++ = a->a[k];
1620: (*mat_ilen)++;
1622: }
1623: }
1624: }
1625: /* Free work space */
1626: ISRestoreIndices(iscol,&icol);
1627: PetscFree(smap);
1628: PetscFree(lens);
1629: }
1630: MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);
1631: MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);
1633: ISRestoreIndices(isrow,&irow);
1634: *B = C;
1635: return(0);
1636: }
1638: /*
1639: */
1642: int MatILUFactor_SeqAIJ(Mat inA,IS row,IS col,MatFactorInfo *info)
1643: {
1644: Mat_SeqAIJ *a = (Mat_SeqAIJ*)inA->data;
1645: int ierr;
1646: Mat outA;
1647: PetscTruth row_identity,col_identity;
1650: if (info->levels != 0) SETERRQ(PETSC_ERR_SUP,"Only levels=0 supported for in-place ilu");
1651: ISIdentity(row,&row_identity);
1652: ISIdentity(col,&col_identity);
1653: if (!row_identity || !col_identity) {
1654: SETERRQ(1,"Row and column permutations must be identity for in-place ILU");
1655: }
1657: outA = inA;
1658: inA->factor = FACTOR_LU;
1659: a->row = row;
1660: a->col = col;
1661: PetscObjectReference((PetscObject)row);
1662: PetscObjectReference((PetscObject)col);
1664: /* Create the inverse permutation so that it can be used in MatLUFactorNumeric() */
1665: if (a->icol) {ISDestroy(a->icol);} /* need to remove old one */
1666: ISInvertPermutation(col,PETSC_DECIDE,&a->icol);
1667: PetscLogObjectParent(inA,a->icol);
1669: if (!a->solve_work) { /* this matrix may have been factored before */
1670: PetscMalloc((inA->m+1)*sizeof(PetscScalar),&a->solve_work);
1671: }
1673: if (!a->diag) {
1674: MatMarkDiagonal_SeqAIJ(inA);
1675: }
1676: MatLUFactorNumeric_SeqAIJ(inA,&outA);
1677: return(0);
1678: }
1680: #include petscblaslapack.h
1683: int MatScale_SeqAIJ(const PetscScalar *alpha,Mat inA)
1684: {
1685: Mat_SeqAIJ *a = (Mat_SeqAIJ*)inA->data;
1686: int one = 1;
1689: BLscal_(&a->nz,(PetscScalar*)alpha,a->a,&one);
1690: PetscLogFlops(a->nz);
1691: return(0);
1692: }
1696: int MatGetSubMatrices_SeqAIJ(Mat A,int n,const IS irow[],const IS icol[],MatReuse scall,Mat *B[])
1697: {
1698: int ierr,i;
1701: if (scall == MAT_INITIAL_MATRIX) {
1702: PetscMalloc((n+1)*sizeof(Mat),B);
1703: }
1705: for (i=0; i<n; i++) {
1706: MatGetSubMatrix_SeqAIJ(A,irow[i],icol[i],PETSC_DECIDE,scall,&(*B)[i]);
1707: }
1708: return(0);
1709: }
1713: int MatGetBlockSize_SeqAIJ(Mat A,int *bs)
1714: {
1716: *bs = 1;
1717: return(0);
1718: }
1722: int MatIncreaseOverlap_SeqAIJ(Mat A,int is_max,IS is[],int ov)
1723: {
1724: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1725: int row,i,j,k,l,m,n,*idx,ierr,*nidx,isz,val;
1726: int start,end,*ai,*aj;
1727: PetscBT table;
1730: m = A->m;
1731: ai = a->i;
1732: aj = a->j;
1734: if (ov < 0) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"illegal negative overlap value used");
1736: PetscMalloc((m+1)*sizeof(int),&nidx);
1737: PetscBTCreate(m,table);
1739: for (i=0; i<is_max; i++) {
1740: /* Initialize the two local arrays */
1741: isz = 0;
1742: PetscBTMemzero(m,table);
1743:
1744: /* Extract the indices, assume there can be duplicate entries */
1745: ISGetIndices(is[i],&idx);
1746: ISGetLocalSize(is[i],&n);
1747:
1748: /* Enter these into the temp arrays. I.e., mark table[row], enter row into new index */
1749: for (j=0; j<n ; ++j){
1750: if(!PetscBTLookupSet(table,idx[j])) { nidx[isz++] = idx[j];}
1751: }
1752: ISRestoreIndices(is[i],&idx);
1753: ISDestroy(is[i]);
1754:
1755: k = 0;
1756: for (j=0; j<ov; j++){ /* for each overlap */
1757: n = isz;
1758: for (; k<n ; k++){ /* do only those rows in nidx[k], which are not done yet */
1759: row = nidx[k];
1760: start = ai[row];
1761: end = ai[row+1];
1762: for (l = start; l<end ; l++){
1763: val = aj[l] ;
1764: if (!PetscBTLookupSet(table,val)) {nidx[isz++] = val;}
1765: }
1766: }
1767: }
1768: ISCreateGeneral(PETSC_COMM_SELF,isz,nidx,(is+i));
1769: }
1770: PetscBTDestroy(table);
1771: PetscFree(nidx);
1772: return(0);
1773: }
1775: /* -------------------------------------------------------------- */
1778: int MatPermute_SeqAIJ(Mat A,IS rowp,IS colp,Mat *B)
1779: {
1780: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1781: PetscScalar *vwork;
1782: int i,ierr,nz,m = A->m,n = A->n,*cwork;
1783: int *row,*col,*cnew,j,*lens;
1784: IS icolp,irowp;
1787: ISInvertPermutation(rowp,PETSC_DECIDE,&irowp);
1788: ISGetIndices(irowp,&row);
1789: ISInvertPermutation(colp,PETSC_DECIDE,&icolp);
1790: ISGetIndices(icolp,&col);
1791:
1792: /* determine lengths of permuted rows */
1793: PetscMalloc((m+1)*sizeof(int),&lens);
1794: for (i=0; i<m; i++) {
1795: lens[row[i]] = a->i[i+1] - a->i[i];
1796: }
1797: MatCreate(A->comm,m,n,m,n,B);
1798: MatSetType(*B,A->type_name);
1799: MatSeqAIJSetPreallocation(*B,0,lens);
1800: PetscFree(lens);
1802: PetscMalloc(n*sizeof(int),&cnew);
1803: for (i=0; i<m; i++) {
1804: MatGetRow(A,i,&nz,&cwork,&vwork);
1805: for (j=0; j<nz; j++) { cnew[j] = col[cwork[j]];}
1806: MatSetValues(*B,1,&row[i],nz,cnew,vwork,INSERT_VALUES);
1807: MatRestoreRow(A,i,&nz,&cwork,&vwork);
1808: }
1809: PetscFree(cnew);
1810: MatAssemblyBegin(*B,MAT_FINAL_ASSEMBLY);
1811: MatAssemblyEnd(*B,MAT_FINAL_ASSEMBLY);
1812: ISRestoreIndices(irowp,&row);
1813: ISRestoreIndices(icolp,&col);
1814: ISDestroy(irowp);
1815: ISDestroy(icolp);
1816: return(0);
1817: }
1821: int MatPrintHelp_SeqAIJ(Mat A)
1822: {
1823: static PetscTruth called = PETSC_FALSE;
1824: MPI_Comm comm = A->comm;
1825: int ierr;
1828: if (called) {return(0);} else called = PETSC_TRUE;
1829: (*PetscHelpPrintf)(comm," Options for MATSEQAIJ and MATMPIAIJ matrix formats (the defaults):\n");
1830: (*PetscHelpPrintf)(comm," -mat_lu_pivotthreshold <threshold>: Set pivoting threshold\n");
1831: (*PetscHelpPrintf)(comm," -mat_aij_oneindex: internal indices begin at 1 instead of the default 0.\n");
1832: (*PetscHelpPrintf)(comm," -mat_aij_no_inode: Do not use inodes\n");
1833: (*PetscHelpPrintf)(comm," -mat_aij_inode_limit <limit>: Set inode limit (max limit=5)\n");
1834: return(0);
1835: }
1839: int MatCopy_SeqAIJ(Mat A,Mat B,MatStructure str)
1840: {
1841: int ierr;
1844: /* If the two matrices have the same copy implementation, use fast copy. */
1845: if (str == SAME_NONZERO_PATTERN && (A->ops->copy == B->ops->copy)) {
1846: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1847: Mat_SeqAIJ *b = (Mat_SeqAIJ*)B->data;
1849: if (a->i[A->m] != b->i[B->m]) {
1850: SETERRQ(1,"Number of nonzeros in two matrices are different");
1851: }
1852: PetscMemcpy(b->a,a->a,(a->i[A->m])*sizeof(PetscScalar));
1853: } else {
1854: MatCopy_Basic(A,B,str);
1855: }
1856: return(0);
1857: }
1861: int MatSetUpPreallocation_SeqAIJ(Mat A)
1862: {
1863: int ierr;
1866: MatSeqAIJSetPreallocation(A,PETSC_DEFAULT,0);
1867: return(0);
1868: }
1872: int MatGetArray_SeqAIJ(Mat A,PetscScalar *array[])
1873: {
1874: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
1876: *array = a->a;
1877: return(0);
1878: }
1882: int MatRestoreArray_SeqAIJ(Mat A,PetscScalar *array[])
1883: {
1885: return(0);
1886: }
1890: int MatFDColoringApply_SeqAIJ(Mat J,MatFDColoring coloring,Vec x1,MatStructure *flag,void *sctx)
1891: {
1892: int (*f)(void *,Vec,Vec,void*) = (int (*)(void *,Vec,Vec,void *))coloring->f;
1893: int k,ierr,N,start,end,l,row,col,srow,**vscaleforrow,m1,m2;
1894: PetscScalar dx,mone = -1.0,*y,*xx,*w3_array;
1895: PetscScalar *vscale_array;
1896: PetscReal epsilon = coloring->error_rel,umin = coloring->umin;
1897: Vec w1,w2,w3;
1898: void *fctx = coloring->fctx;
1899: PetscTruth flg;
1902: if (!coloring->w1) {
1903: VecDuplicate(x1,&coloring->w1);
1904: PetscLogObjectParent(coloring,coloring->w1);
1905: VecDuplicate(x1,&coloring->w2);
1906: PetscLogObjectParent(coloring,coloring->w2);
1907: VecDuplicate(x1,&coloring->w3);
1908: PetscLogObjectParent(coloring,coloring->w3);
1909: }
1910: w1 = coloring->w1; w2 = coloring->w2; w3 = coloring->w3;
1912: MatSetUnfactored(J);
1913: PetscOptionsHasName(coloring->prefix,"-mat_fd_coloring_dont_rezero",&flg);
1914: if (flg) {
1915: PetscLogInfo(coloring,"MatFDColoringApply_SeqAIJ: Not calling MatZeroEntries()\n");
1916: } else {
1917: MatZeroEntries(J);
1918: }
1920: VecGetOwnershipRange(x1,&start,&end);
1921: VecGetSize(x1,&N);
1923: /*
1924: This is a horrible, horrible, hack. See DMMGComputeJacobian_Multigrid() it inproperly sets
1925: coloring->F for the coarser grids from the finest
1926: */
1927: if (coloring->F) {
1928: VecGetLocalSize(coloring->F,&m1);
1929: VecGetLocalSize(w1,&m2);
1930: if (m1 != m2) {
1931: coloring->F = 0;
1932: }
1933: }
1935: if (coloring->F) {
1936: w1 = coloring->F;
1937: coloring->F = 0;
1938: } else {
1939: PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);
1940: (*f)(sctx,x1,w1,fctx);
1941: PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);
1942: }
1944: /*
1945: Compute all the scale factors and share with other processors
1946: */
1947: VecGetArray(x1,&xx);xx = xx - start;
1948: VecGetArray(coloring->vscale,&vscale_array);vscale_array = vscale_array - start;
1949: for (k=0; k<coloring->ncolors; k++) {
1950: /*
1951: Loop over each column associated with color adding the
1952: perturbation to the vector w3.
1953: */
1954: for (l=0; l<coloring->ncolumns[k]; l++) {
1955: col = coloring->columns[k][l]; /* column of the matrix we are probing for */
1956: dx = xx[col];
1957: if (dx == 0.0) dx = 1.0;
1958: #if !defined(PETSC_USE_COMPLEX)
1959: if (dx < umin && dx >= 0.0) dx = umin;
1960: else if (dx < 0.0 && dx > -umin) dx = -umin;
1961: #else
1962: if (PetscAbsScalar(dx) < umin && PetscRealPart(dx) >= 0.0) dx = umin;
1963: else if (PetscRealPart(dx) < 0.0 && PetscAbsScalar(dx) < umin) dx = -umin;
1964: #endif
1965: dx *= epsilon;
1966: vscale_array[col] = 1.0/dx;
1967: }
1968: }
1969: vscale_array = vscale_array + start;VecRestoreArray(coloring->vscale,&vscale_array);
1970: VecGhostUpdateBegin(coloring->vscale,INSERT_VALUES,SCATTER_FORWARD);
1971: VecGhostUpdateEnd(coloring->vscale,INSERT_VALUES,SCATTER_FORWARD);
1973: /* VecView(coloring->vscale,PETSC_VIEWER_STDOUT_WORLD);
1974: VecView(x1,PETSC_VIEWER_STDOUT_WORLD);*/
1976: if (coloring->vscaleforrow) vscaleforrow = coloring->vscaleforrow;
1977: else vscaleforrow = coloring->columnsforrow;
1979: VecGetArray(coloring->vscale,&vscale_array);
1980: /*
1981: Loop over each color
1982: */
1983: for (k=0; k<coloring->ncolors; k++) {
1984: coloring->currentcolor = k;
1985: VecCopy(x1,w3);
1986: VecGetArray(w3,&w3_array);w3_array = w3_array - start;
1987: /*
1988: Loop over each column associated with color adding the
1989: perturbation to the vector w3.
1990: */
1991: for (l=0; l<coloring->ncolumns[k]; l++) {
1992: col = coloring->columns[k][l]; /* column of the matrix we are probing for */
1993: dx = xx[col];
1994: if (dx == 0.0) dx = 1.0;
1995: #if !defined(PETSC_USE_COMPLEX)
1996: if (dx < umin && dx >= 0.0) dx = umin;
1997: else if (dx < 0.0 && dx > -umin) dx = -umin;
1998: #else
1999: if (PetscAbsScalar(dx) < umin && PetscRealPart(dx) >= 0.0) dx = umin;
2000: else if (PetscRealPart(dx) < 0.0 && PetscAbsScalar(dx) < umin) dx = -umin;
2001: #endif
2002: dx *= epsilon;
2003: if (!PetscAbsScalar(dx)) SETERRQ(1,"Computed 0 differencing parameter");
2004: w3_array[col] += dx;
2005: }
2006: w3_array = w3_array + start; VecRestoreArray(w3,&w3_array);
2008: /*
2009: Evaluate function at x1 + dx (here dx is a vector of perturbations)
2010: */
2012: PetscLogEventBegin(MAT_FDColoringFunction,0,0,0,0);
2013: (*f)(sctx,w3,w2,fctx);
2014: PetscLogEventEnd(MAT_FDColoringFunction,0,0,0,0);
2015: VecAXPY(&mone,w1,w2);
2017: /*
2018: Loop over rows of vector, putting results into Jacobian matrix
2019: */
2020: VecGetArray(w2,&y);
2021: for (l=0; l<coloring->nrows[k]; l++) {
2022: row = coloring->rows[k][l];
2023: col = coloring->columnsforrow[k][l];
2024: y[row] *= vscale_array[vscaleforrow[k][l]];
2025: srow = row + start;
2026: MatSetValues_SeqAIJ(J,1,&srow,1,&col,y+row,INSERT_VALUES);
2027: }
2028: VecRestoreArray(w2,&y);
2029: }
2030: coloring->currentcolor = k;
2031: VecRestoreArray(coloring->vscale,&vscale_array);
2032: xx = xx + start; VecRestoreArray(x1,&xx);
2033: MatAssemblyBegin(J,MAT_FINAL_ASSEMBLY);
2034: MatAssemblyEnd(J,MAT_FINAL_ASSEMBLY);
2035: return(0);
2036: }
2038: #include petscblaslapack.h
2041: int MatAXPY_SeqAIJ(const PetscScalar a[],Mat X,Mat Y,MatStructure str)
2042: {
2043: int ierr,one=1,i;
2044: Mat_SeqAIJ *x = (Mat_SeqAIJ *)X->data,*y = (Mat_SeqAIJ *)Y->data;
2047: if (str == SAME_NONZERO_PATTERN) {
2048: BLaxpy_(&x->nz,(PetscScalar*)a,x->a,&one,y->a,&one);
2049: } else if (str == SUBSET_NONZERO_PATTERN) { /* nonzeros of X is a subset of Y's */
2050: if (y->xtoy && y->XtoY != X) {
2051: PetscFree(y->xtoy);
2052: MatDestroy(y->XtoY);
2053: }
2054: if (!y->xtoy) { /* get xtoy */
2055: MatAXPYGetxtoy_Private(X->m,x->i,x->j,PETSC_NULL, y->i,y->j,PETSC_NULL, &y->xtoy);
2056: y->XtoY = X;
2057: }
2058: for (i=0; i<x->nz; i++) y->a[y->xtoy[i]] += (*a)*(x->a[i]);
2059: PetscLogInfo(0,"MatAXPY_SeqAIJ: ratio of nnz(X)/nnz(Y): %d/%d = %g\n",x->nz,y->nz,(PetscReal)(x->nz)/y->nz);
2060: } else {
2061: MatAXPY_Basic(a,X,Y,str);
2062: }
2063: return(0);
2064: }
2066: /* -------------------------------------------------------------------*/
2067: static struct _MatOps MatOps_Values = {MatSetValues_SeqAIJ,
2068: MatGetRow_SeqAIJ,
2069: MatRestoreRow_SeqAIJ,
2070: MatMult_SeqAIJ,
2071: /* 4*/ MatMultAdd_SeqAIJ,
2072: MatMultTranspose_SeqAIJ,
2073: MatMultTransposeAdd_SeqAIJ,
2074: MatSolve_SeqAIJ,
2075: MatSolveAdd_SeqAIJ,
2076: MatSolveTranspose_SeqAIJ,
2077: /*10*/ MatSolveTransposeAdd_SeqAIJ,
2078: MatLUFactor_SeqAIJ,
2079: 0,
2080: MatRelax_SeqAIJ,
2081: MatTranspose_SeqAIJ,
2082: /*15*/ MatGetInfo_SeqAIJ,
2083: MatEqual_SeqAIJ,
2084: MatGetDiagonal_SeqAIJ,
2085: MatDiagonalScale_SeqAIJ,
2086: MatNorm_SeqAIJ,
2087: /*20*/ 0,
2088: MatAssemblyEnd_SeqAIJ,
2089: MatCompress_SeqAIJ,
2090: MatSetOption_SeqAIJ,
2091: MatZeroEntries_SeqAIJ,
2092: /*25*/ MatZeroRows_SeqAIJ,
2093: MatLUFactorSymbolic_SeqAIJ,
2094: MatLUFactorNumeric_SeqAIJ,
2095: MatCholeskyFactorSymbolic_SeqAIJ,
2096: MatCholeskyFactorNumeric_SeqAIJ,
2097: /*30*/ MatSetUpPreallocation_SeqAIJ,
2098: MatILUFactorSymbolic_SeqAIJ,
2099: MatICCFactorSymbolic_SeqAIJ,
2100: MatGetArray_SeqAIJ,
2101: MatRestoreArray_SeqAIJ,
2102: /*35*/ MatDuplicate_SeqAIJ,
2103: 0,
2104: 0,
2105: MatILUFactor_SeqAIJ,
2106: 0,
2107: /*40*/ MatAXPY_SeqAIJ,
2108: MatGetSubMatrices_SeqAIJ,
2109: MatIncreaseOverlap_SeqAIJ,
2110: MatGetValues_SeqAIJ,
2111: MatCopy_SeqAIJ,
2112: /*45*/ MatPrintHelp_SeqAIJ,
2113: MatScale_SeqAIJ,
2114: 0,
2115: 0,
2116: MatILUDTFactor_SeqAIJ,
2117: /*50*/ MatGetBlockSize_SeqAIJ,
2118: MatGetRowIJ_SeqAIJ,
2119: MatRestoreRowIJ_SeqAIJ,
2120: MatGetColumnIJ_SeqAIJ,
2121: MatRestoreColumnIJ_SeqAIJ,
2122: /*55*/ MatFDColoringCreate_SeqAIJ,
2123: 0,
2124: 0,
2125: MatPermute_SeqAIJ,
2126: 0,
2127: /*60*/ 0,
2128: MatDestroy_SeqAIJ,
2129: MatView_SeqAIJ,
2130: MatGetPetscMaps_Petsc,
2131: 0,
2132: /*65*/ 0,
2133: 0,
2134: 0,
2135: 0,
2136: 0,
2137: /*70*/ 0,
2138: 0,
2139: MatSetColoring_SeqAIJ,
2140: MatSetValuesAdic_SeqAIJ,
2141: MatSetValuesAdifor_SeqAIJ,
2142: /*75*/ MatFDColoringApply_SeqAIJ,
2143: 0,
2144: 0,
2145: 0,
2146: 0,
2147: /*80*/ 0,
2148: 0,
2149: 0,
2150: 0,
2151: /*85*/ MatLoad_SeqAIJ,
2152: MatIsSymmetric_SeqAIJ,
2153: };
2155: EXTERN_C_BEGIN
2158: int MatSeqAIJSetColumnIndices_SeqAIJ(Mat mat,int *indices)
2159: {
2160: Mat_SeqAIJ *aij = (Mat_SeqAIJ *)mat->data;
2161: int i,nz,n;
2165: nz = aij->maxnz;
2166: n = mat->n;
2167: for (i=0; i<nz; i++) {
2168: aij->j[i] = indices[i];
2169: }
2170: aij->nz = nz;
2171: for (i=0; i<n; i++) {
2172: aij->ilen[i] = aij->imax[i];
2173: }
2175: return(0);
2176: }
2177: EXTERN_C_END
2181: /*@
2182: MatSeqAIJSetColumnIndices - Set the column indices for all the rows
2183: in the matrix.
2185: Input Parameters:
2186: + mat - the SeqAIJ matrix
2187: - indices - the column indices
2189: Level: advanced
2191: Notes:
2192: This can be called if you have precomputed the nonzero structure of the
2193: matrix and want to provide it to the matrix object to improve the performance
2194: of the MatSetValues() operation.
2196: You MUST have set the correct numbers of nonzeros per row in the call to
2197: MatCreateSeqAIJ().
2199: MUST be called before any calls to MatSetValues();
2201: The indices should start with zero, not one.
2203: @*/
2204: int MatSeqAIJSetColumnIndices(Mat mat,int *indices)
2205: {
2206: int ierr,(*f)(Mat,int *);
2211: PetscObjectQueryFunction((PetscObject)mat,"MatSeqAIJSetColumnIndices_C",(void (**)(void))&f);
2212: if (f) {
2213: (*f)(mat,indices);
2214: } else {
2215: SETERRQ(1,"Wrong type of matrix to set column indices");
2216: }
2217: return(0);
2218: }
2220: /* ----------------------------------------------------------------------------------------*/
2222: EXTERN_C_BEGIN
2225: int MatStoreValues_SeqAIJ(Mat mat)
2226: {
2227: Mat_SeqAIJ *aij = (Mat_SeqAIJ *)mat->data;
2228: size_t nz = aij->i[mat->m],ierr;
2231: if (aij->nonew != 1) {
2232: SETERRQ(1,"Must call MatSetOption(A,MAT_NO_NEW_NONZERO_LOCATIONS);first");
2233: }
2235: /* allocate space for values if not already there */
2236: if (!aij->saved_values) {
2237: PetscMalloc((nz+1)*sizeof(PetscScalar),&aij->saved_values);
2238: }
2240: /* copy values over */
2241: PetscMemcpy(aij->saved_values,aij->a,nz*sizeof(PetscScalar));
2242: return(0);
2243: }
2244: EXTERN_C_END
2248: /*@
2249: MatStoreValues - Stashes a copy of the matrix values; this allows, for
2250: example, reuse of the linear part of a Jacobian, while recomputing the
2251: nonlinear portion.
2253: Collect on Mat
2255: Input Parameters:
2256: . mat - the matrix (currently on AIJ matrices support this option)
2258: Level: advanced
2260: Common Usage, with SNESSolve():
2261: $ Create Jacobian matrix
2262: $ Set linear terms into matrix
2263: $ Apply boundary conditions to matrix, at this time matrix must have
2264: $ final nonzero structure (i.e. setting the nonlinear terms and applying
2265: $ boundary conditions again will not change the nonzero structure
2266: $ MatSetOption(mat,MAT_NO_NEW_NONZERO_LOCATIONS);
2267: $ MatStoreValues(mat);
2268: $ Call SNESSetJacobian() with matrix
2269: $ In your Jacobian routine
2270: $ MatRetrieveValues(mat);
2271: $ Set nonlinear terms in matrix
2272:
2273: Common Usage without SNESSolve(), i.e. when you handle nonlinear solve yourself:
2274: $ // build linear portion of Jacobian
2275: $ MatSetOption(mat,MAT_NO_NEW_NONZERO_LOCATIONS);
2276: $ MatStoreValues(mat);
2277: $ loop over nonlinear iterations
2278: $ MatRetrieveValues(mat);
2279: $ // call MatSetValues(mat,...) to set nonliner portion of Jacobian
2280: $ // call MatAssemblyBegin/End() on matrix
2281: $ Solve linear system with Jacobian
2282: $ endloop
2284: Notes:
2285: Matrix must already be assemblied before calling this routine
2286: Must set the matrix option MatSetOption(mat,MAT_NO_NEW_NONZERO_LOCATIONS); before
2287: calling this routine.
2289: .seealso: MatRetrieveValues()
2291: @*/
2292: int MatStoreValues(Mat mat)
2293: {
2294: int ierr,(*f)(Mat);
2298: if (!mat->assembled) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix");
2299: if (mat->factor) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
2301: PetscObjectQueryFunction((PetscObject)mat,"MatStoreValues_C",(void (**)(void))&f);
2302: if (f) {
2303: (*f)(mat);
2304: } else {
2305: SETERRQ(1,"Wrong type of matrix to store values");
2306: }
2307: return(0);
2308: }
2310: EXTERN_C_BEGIN
2313: int MatRetrieveValues_SeqAIJ(Mat mat)
2314: {
2315: Mat_SeqAIJ *aij = (Mat_SeqAIJ *)mat->data;
2316: int nz = aij->i[mat->m],ierr;
2319: if (aij->nonew != 1) {
2320: SETERRQ(1,"Must call MatSetOption(A,MAT_NO_NEW_NONZERO_LOCATIONS);first");
2321: }
2322: if (!aij->saved_values) {
2323: SETERRQ(1,"Must call MatStoreValues(A);first");
2324: }
2326: /* copy values over */
2327: PetscMemcpy(aij->a,aij->saved_values,nz*sizeof(PetscScalar));
2328: return(0);
2329: }
2330: EXTERN_C_END
2334: /*@
2335: MatRetrieveValues - Retrieves the copy of the matrix values; this allows, for
2336: example, reuse of the linear part of a Jacobian, while recomputing the
2337: nonlinear portion.
2339: Collect on Mat
2341: Input Parameters:
2342: . mat - the matrix (currently on AIJ matrices support this option)
2344: Level: advanced
2346: .seealso: MatStoreValues()
2348: @*/
2349: int MatRetrieveValues(Mat mat)
2350: {
2351: int ierr,(*f)(Mat);
2355: if (!mat->assembled) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix");
2356: if (mat->factor) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
2358: PetscObjectQueryFunction((PetscObject)mat,"MatRetrieveValues_C",(void (**)(void))&f);
2359: if (f) {
2360: (*f)(mat);
2361: } else {
2362: SETERRQ(1,"Wrong type of matrix to retrieve values");
2363: }
2364: return(0);
2365: }
2368: /* --------------------------------------------------------------------------------*/
2371: /*@C
2372: MatCreateSeqAIJ - Creates a sparse matrix in AIJ (compressed row) format
2373: (the default parallel PETSc format). For good matrix assembly performance
2374: the user should preallocate the matrix storage by setting the parameter nz
2375: (or the array nnz). By setting these parameters accurately, performance
2376: during matrix assembly can be increased by more than a factor of 50.
2378: Collective on MPI_Comm
2380: Input Parameters:
2381: + comm - MPI communicator, set to PETSC_COMM_SELF
2382: . m - number of rows
2383: . n - number of columns
2384: . nz - number of nonzeros per row (same for all rows)
2385: - nnz - array containing the number of nonzeros in the various rows
2386: (possibly different for each row) or PETSC_NULL
2388: Output Parameter:
2389: . A - the matrix
2391: Notes:
2392: The AIJ format (also called the Yale sparse matrix format or
2393: compressed row storage), is fully compatible with standard Fortran 77
2394: storage. That is, the stored row and column indices can begin at
2395: either one (as in Fortran) or zero. See the users' manual for details.
2397: Specify the preallocated storage with either nz or nnz (not both).
2398: Set nz=PETSC_DEFAULT and nnz=PETSC_NULL for PETSc to control dynamic memory
2399: allocation. For large problems you MUST preallocate memory or you
2400: will get TERRIBLE performance, see the users' manual chapter on matrices.
2402: By default, this format uses inodes (identical nodes) when possible, to
2403: improve numerical efficiency of matrix-vector products and solves. We
2404: search for consecutive rows with the same nonzero structure, thereby
2405: reusing matrix information to achieve increased efficiency.
2407: Options Database Keys:
2408: + -mat_aij_no_inode - Do not use inodes
2409: . -mat_aij_inode_limit <limit> - Sets inode limit (max limit=5)
2410: - -mat_aij_oneindex - Internally use indexing starting at 1
2411: rather than 0. Note that when calling MatSetValues(),
2412: the user still MUST index entries starting at 0!
2414: Level: intermediate
2416: .seealso: MatCreate(), MatCreateMPIAIJ(), MatSetValues(), MatSeqAIJSetColumnIndices(), MatCreateSeqAIJWithArrays()
2418: @*/
2419: int MatCreateSeqAIJ(MPI_Comm comm,int m,int n,int nz,const int nnz[],Mat *A)
2420: {
2424: MatCreate(comm,m,n,m,n,A);
2425: MatSetType(*A,MATSEQAIJ);
2426: MatSeqAIJSetPreallocation(*A,nz,nnz);
2427: return(0);
2428: }
2430: #define SKIP_ALLOCATION -4
2434: /*@C
2435: MatSeqAIJSetPreallocation - For good matrix assembly performance
2436: the user should preallocate the matrix storage by setting the parameter nz
2437: (or the array nnz). By setting these parameters accurately, performance
2438: during matrix assembly can be increased by more than a factor of 50.
2440: Collective on MPI_Comm
2442: Input Parameters:
2443: + comm - MPI communicator, set to PETSC_COMM_SELF
2444: . m - number of rows
2445: . n - number of columns
2446: . nz - number of nonzeros per row (same for all rows)
2447: - nnz - array containing the number of nonzeros in the various rows
2448: (possibly different for each row) or PETSC_NULL
2450: Output Parameter:
2451: . A - the matrix
2453: Notes:
2454: The AIJ format (also called the Yale sparse matrix format or
2455: compressed row storage), is fully compatible with standard Fortran 77
2456: storage. That is, the stored row and column indices can begin at
2457: either one (as in Fortran) or zero. See the users' manual for details.
2459: Specify the preallocated storage with either nz or nnz (not both).
2460: Set nz=PETSC_DEFAULT and nnz=PETSC_NULL for PETSc to control dynamic memory
2461: allocation. For large problems you MUST preallocate memory or you
2462: will get TERRIBLE performance, see the users' manual chapter on matrices.
2464: By default, this format uses inodes (identical nodes) when possible, to
2465: improve numerical efficiency of matrix-vector products and solves. We
2466: search for consecutive rows with the same nonzero structure, thereby
2467: reusing matrix information to achieve increased efficiency.
2469: Options Database Keys:
2470: + -mat_aij_no_inode - Do not use inodes
2471: . -mat_aij_inode_limit <limit> - Sets inode limit (max limit=5)
2472: - -mat_aij_oneindex - Internally use indexing starting at 1
2473: rather than 0. Note that when calling MatSetValues(),
2474: the user still MUST index entries starting at 0!
2476: Level: intermediate
2478: .seealso: MatCreate(), MatCreateMPIAIJ(), MatSetValues(), MatSeqAIJSetColumnIndices(), MatCreateSeqAIJWithArrays()
2480: @*/
2481: int MatSeqAIJSetPreallocation(Mat B,int nz,const int nnz[])
2482: {
2483: int ierr,(*f)(Mat,int,const int[]);
2486: PetscObjectQueryFunction((PetscObject)B,"MatSeqAIJSetPreallocation_C",(void (**)(void))&f);
2487: if (f) {
2488: (*f)(B,nz,nnz);
2489: }
2490: return(0);
2491: }
2493: EXTERN_C_BEGIN
2496: int MatSeqAIJSetPreallocation_SeqAIJ(Mat B,int nz,int *nnz)
2497: {
2498: Mat_SeqAIJ *b;
2499: size_t len = 0;
2500: PetscTruth skipallocation = PETSC_FALSE;
2501: int i,ierr;
2504:
2505: if (nz == SKIP_ALLOCATION) {
2506: skipallocation = PETSC_TRUE;
2507: nz = 0;
2508: }
2510: if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 5;
2511: if (nz < 0) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"nz cannot be less than 0: value %d",nz);
2512: if (nnz) {
2513: for (i=0; i<B->m; i++) {
2514: if (nnz[i] < 0) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"nnz cannot be less than 0: local row %d value %d",i,nnz[i]);
2515: if (nnz[i] > B->n) SETERRQ3(PETSC_ERR_ARG_OUTOFRANGE,"nnz cannot be greater than row length: local row %d value %d rowlength %d",i,nnz[i],B->n);
2516: }
2517: }
2519: B->preallocated = PETSC_TRUE;
2520: b = (Mat_SeqAIJ*)B->data;
2522: PetscMalloc((B->m+1)*sizeof(int),&b->imax);
2523: if (!nnz) {
2524: if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 10;
2525: else if (nz <= 0) nz = 1;
2526: for (i=0; i<B->m; i++) b->imax[i] = nz;
2527: nz = nz*B->m;
2528: } else {
2529: nz = 0;
2530: for (i=0; i<B->m; i++) {b->imax[i] = nnz[i]; nz += nnz[i];}
2531: }
2533: if (!skipallocation) {
2534: /* allocate the matrix space */
2535: len = ((size_t) nz)*(sizeof(int) + sizeof(PetscScalar)) + (B->m+1)*sizeof(int);
2536: PetscMalloc(len,&b->a);
2537: b->j = (int*)(b->a + nz);
2538: PetscMemzero(b->j,nz*sizeof(int));
2539: b->i = b->j + nz;
2540: b->i[0] = 0;
2541: for (i=1; i<B->m+1; i++) {
2542: b->i[i] = b->i[i-1] + b->imax[i-1];
2543: }
2544: b->singlemalloc = PETSC_TRUE;
2545: b->freedata = PETSC_TRUE;
2546: } else {
2547: b->freedata = PETSC_FALSE;
2548: }
2550: /* b->ilen will count nonzeros in each row so far. */
2551: PetscMalloc((B->m+1)*sizeof(int),&b->ilen);
2552: PetscLogObjectMemory(B,len+2*(B->m+1)*sizeof(int)+sizeof(struct _p_Mat)+sizeof(Mat_SeqAIJ));
2553: for (i=0; i<B->m; i++) { b->ilen[i] = 0;}
2555: b->nz = 0;
2556: b->maxnz = nz;
2557: B->info.nz_unneeded = (double)b->maxnz;
2558: return(0);
2559: }
2560: EXTERN_C_END
2562: /*MC
2563: MATSEQAIJ - MATSEQAIJ = "seqaij" - A matrix type to be used for sequential sparse matrices,
2564: based on compressed sparse row format.
2566: Options Database Keys:
2567: . -mat_type seqaij - sets the matrix type to "seqaij" during a call to MatSetFromOptions()
2569: Level: beginner
2571: .seealso: MatCreateSeqAIJ
2572: M*/
2574: EXTERN_C_BEGIN
2577: int MatCreate_SeqAIJ(Mat B)
2578: {
2579: Mat_SeqAIJ *b;
2580: int ierr,size;
2583: MPI_Comm_size(B->comm,&size);
2584: if (size > 1) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Comm must be of size 1");
2586: B->m = B->M = PetscMax(B->m,B->M);
2587: B->n = B->N = PetscMax(B->n,B->N);
2589: PetscNew(Mat_SeqAIJ,&b);
2590: B->data = (void*)b;
2591: PetscMemzero(b,sizeof(Mat_SeqAIJ));
2592: PetscMemcpy(B->ops,&MatOps_Values,sizeof(struct _MatOps));
2593: B->factor = 0;
2594: B->lupivotthreshold = 1.0;
2595: B->mapping = 0;
2596: PetscOptionsGetReal(B->prefix,"-mat_lu_pivotthreshold",&B->lupivotthreshold,PETSC_NULL);
2597: PetscOptionsHasName(B->prefix,"-pc_ilu_preserve_row_sums",&b->ilu_preserve_row_sums);
2598: b->row = 0;
2599: b->col = 0;
2600: b->icol = 0;
2601: b->reallocs = 0;
2602:
2603: PetscMapCreateMPI(B->comm,B->m,B->m,&B->rmap);
2604: PetscMapCreateMPI(B->comm,B->n,B->n,&B->cmap);
2606: b->sorted = PETSC_FALSE;
2607: b->ignorezeroentries = PETSC_FALSE;
2608: b->roworiented = PETSC_TRUE;
2609: b->nonew = 0;
2610: b->diag = 0;
2611: b->solve_work = 0;
2612: B->spptr = 0;
2613: b->inode.use = PETSC_TRUE;
2614: b->inode.node_count = 0;
2615: b->inode.size = 0;
2616: b->inode.limit = 5;
2617: b->inode.max_limit = 5;
2618: b->saved_values = 0;
2619: b->idiag = 0;
2620: b->ssor = 0;
2621: b->keepzeroedrows = PETSC_FALSE;
2622: b->xtoy = 0;
2623: b->XtoY = 0;
2625: PetscObjectChangeTypeName((PetscObject)B,MATSEQAIJ);
2627: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatSeqAIJSetColumnIndices_C",
2628: "MatSeqAIJSetColumnIndices_SeqAIJ",
2629: MatSeqAIJSetColumnIndices_SeqAIJ);
2630: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatStoreValues_C",
2631: "MatStoreValues_SeqAIJ",
2632: MatStoreValues_SeqAIJ);
2633: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatRetrieveValues_C",
2634: "MatRetrieveValues_SeqAIJ",
2635: MatRetrieveValues_SeqAIJ);
2636: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatConvert_seqaij_seqsbaij_C",
2637: "MatConvert_SeqAIJ_SeqSBAIJ",
2638: MatConvert_SeqAIJ_SeqSBAIJ);
2639: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatConvert_seqaij_seqbaij_C",
2640: "MatConvert_SeqAIJ_SeqBAIJ",
2641: MatConvert_SeqAIJ_SeqBAIJ);
2642: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatIsTranspose_C",
2643: "MatIsTranspose_SeqAIJ",
2644: MatIsTranspose_SeqAIJ);
2645: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatSeqAIJSetPreallocation_C",
2646: "MatSeqAIJSetPreallocation_SeqAIJ",
2647: MatSeqAIJSetPreallocation_SeqAIJ);
2648: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatReorderForNonzeroDiagonal_C",
2649: "MatReorderForNonzeroDiagonal_SeqAIJ",
2650: MatReorderForNonzeroDiagonal_SeqAIJ);
2651: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatAdjustForInodes_C",
2652: "MatAdjustForInodes_SeqAIJ",
2653: MatAdjustForInodes_SeqAIJ);
2654: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatSeqAIJGetInodeSizes_C",
2655: "MatSeqAIJGetInodeSizes_SeqAIJ",
2656: MatSeqAIJGetInodeSizes_SeqAIJ);
2657: RegisterApplyPtAPRoutines_Private(B);
2658: return(0);
2659: }
2660: EXTERN_C_END
2664: int MatDuplicate_SeqAIJ(Mat A,MatDuplicateOption cpvalues,Mat *B)
2665: {
2666: Mat C;
2667: Mat_SeqAIJ *c,*a = (Mat_SeqAIJ*)A->data;
2668: int i,m = A->m,ierr;
2669: size_t len;
2672: *B = 0;
2673: MatCreate(A->comm,A->m,A->n,A->m,A->n,&C);
2674: MatSetType(C,A->type_name);
2675: c = (Mat_SeqAIJ*)C->data;
2677: C->factor = A->factor;
2678: c->row = 0;
2679: c->col = 0;
2680: c->icol = 0;
2681: c->keepzeroedrows = a->keepzeroedrows;
2682: C->assembled = PETSC_TRUE;
2684: C->M = A->m;
2685: C->N = A->n;
2687: PetscMalloc((m+1)*sizeof(int),&c->imax);
2688: PetscMalloc((m+1)*sizeof(int),&c->ilen);
2689: for (i=0; i<m; i++) {
2690: c->imax[i] = a->imax[i];
2691: c->ilen[i] = a->ilen[i];
2692: }
2694: /* allocate the matrix space */
2695: c->singlemalloc = PETSC_TRUE;
2696: len = ((size_t) (m+1))*sizeof(int)+(a->i[m])*(sizeof(PetscScalar)+sizeof(int));
2697: PetscMalloc(len,&c->a);
2698: c->j = (int*)(c->a + a->i[m] );
2699: c->i = c->j + a->i[m];
2700: PetscMemcpy(c->i,a->i,(m+1)*sizeof(int));
2701: if (m > 0) {
2702: PetscMemcpy(c->j,a->j,(a->i[m])*sizeof(int));
2703: if (cpvalues == MAT_COPY_VALUES) {
2704: PetscMemcpy(c->a,a->a,(a->i[m])*sizeof(PetscScalar));
2705: } else {
2706: PetscMemzero(c->a,(a->i[m])*sizeof(PetscScalar));
2707: }
2708: }
2710: PetscLogObjectMemory(C,len+2*(m+1)*sizeof(int)+sizeof(struct _p_Mat)+sizeof(Mat_SeqAIJ));
2711: c->sorted = a->sorted;
2712: c->roworiented = a->roworiented;
2713: c->nonew = a->nonew;
2714: c->ilu_preserve_row_sums = a->ilu_preserve_row_sums;
2715: c->saved_values = 0;
2716: c->idiag = 0;
2717: c->ssor = 0;
2718: c->ignorezeroentries = a->ignorezeroentries;
2719: c->freedata = PETSC_TRUE;
2721: if (a->diag) {
2722: PetscMalloc((m+1)*sizeof(int),&c->diag);
2723: PetscLogObjectMemory(C,(m+1)*sizeof(int));
2724: for (i=0; i<m; i++) {
2725: c->diag[i] = a->diag[i];
2726: }
2727: } else c->diag = 0;
2728: c->inode.use = a->inode.use;
2729: c->inode.limit = a->inode.limit;
2730: c->inode.max_limit = a->inode.max_limit;
2731: if (a->inode.size){
2732: PetscMalloc((m+1)*sizeof(int),&c->inode.size);
2733: c->inode.node_count = a->inode.node_count;
2734: PetscMemcpy(c->inode.size,a->inode.size,(m+1)*sizeof(int));
2735: } else {
2736: c->inode.size = 0;
2737: c->inode.node_count = 0;
2738: }
2739: c->nz = a->nz;
2740: c->maxnz = a->maxnz;
2741: c->solve_work = 0;
2742: C->spptr = 0; /* Dangerous -I'm throwing away a->spptr */
2743: C->preallocated = PETSC_TRUE;
2745: *B = C;
2746: PetscFListDuplicate(A->qlist,&C->qlist);
2747: return(0);
2748: }
2752: int MatLoad_SeqAIJ(PetscViewer viewer,const MatType type,Mat *A)
2753: {
2754: Mat_SeqAIJ *a;
2755: Mat B;
2756: int i,nz,ierr,fd,header[4],size,*rowlengths = 0,M,N;
2757: MPI_Comm comm;
2758:
2760: PetscObjectGetComm((PetscObject)viewer,&comm);
2761: MPI_Comm_size(comm,&size);
2762: if (size > 1) SETERRQ(PETSC_ERR_ARG_SIZ,"view must have one processor");
2763: PetscViewerBinaryGetDescriptor(viewer,&fd);
2764: PetscBinaryRead(fd,header,4,PETSC_INT);
2765: if (header[0] != MAT_FILE_COOKIE) SETERRQ(PETSC_ERR_FILE_UNEXPECTED,"not matrix object in file");
2766: M = header[1]; N = header[2]; nz = header[3];
2768: if (nz < 0) {
2769: SETERRQ(PETSC_ERR_FILE_UNEXPECTED,"Matrix stored in special format on disk,cannot load as SeqAIJ");
2770: }
2772: /* read in row lengths */
2773: PetscMalloc(M*sizeof(int),&rowlengths);
2774: PetscBinaryRead(fd,rowlengths,M,PETSC_INT);
2776: /* create our matrix */
2777: MatCreate(comm,PETSC_DECIDE,PETSC_DECIDE,M,N,&B);
2778: MatSetType(B,type);
2779: MatSeqAIJSetPreallocation(B,0,rowlengths);
2780: a = (Mat_SeqAIJ*)B->data;
2782: /* read in column indices and adjust for Fortran indexing*/
2783: PetscBinaryRead(fd,a->j,nz,PETSC_INT);
2785: /* read in nonzero values */
2786: PetscBinaryRead(fd,a->a,nz,PETSC_SCALAR);
2788: /* set matrix "i" values */
2789: a->i[0] = 0;
2790: for (i=1; i<= M; i++) {
2791: a->i[i] = a->i[i-1] + rowlengths[i-1];
2792: a->ilen[i-1] = rowlengths[i-1];
2793: }
2794: PetscFree(rowlengths);
2796: MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);
2797: MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);
2798: *A = B;
2799: return(0);
2800: }
2804: int MatEqual_SeqAIJ(Mat A,Mat B,PetscTruth* flg)
2805: {
2806: Mat_SeqAIJ *a = (Mat_SeqAIJ *)A->data,*b = (Mat_SeqAIJ *)B->data;
2807: int ierr;
2810: /* If the matrix dimensions are not equal,or no of nonzeros */
2811: if ((A->m != B->m) || (A->n != B->n) ||(a->nz != b->nz)) {
2812: *flg = PETSC_FALSE;
2813: return(0);
2814: }
2815:
2816: /* if the a->i are the same */
2817: PetscMemcmp(a->i,b->i,(A->m+1)*sizeof(int),flg);
2818: if (*flg == PETSC_FALSE) return(0);
2819:
2820: /* if a->j are the same */
2821: PetscMemcmp(a->j,b->j,(a->nz)*sizeof(int),flg);
2822: if (*flg == PETSC_FALSE) return(0);
2823:
2824: /* if a->a are the same */
2825: PetscMemcmp(a->a,b->a,(a->nz)*sizeof(PetscScalar),flg);
2827: return(0);
2828:
2829: }
2833: /*@C
2834: MatCreateSeqAIJWithArrays - Creates an sequential AIJ matrix using matrix elements (in CSR format)
2835: provided by the user.
2837: Coolective on MPI_Comm
2839: Input Parameters:
2840: + comm - must be an MPI communicator of size 1
2841: . m - number of rows
2842: . n - number of columns
2843: . i - row indices
2844: . j - column indices
2845: - a - matrix values
2847: Output Parameter:
2848: . mat - the matrix
2850: Level: intermediate
2852: Notes:
2853: The i, j, and a arrays are not copied by this routine, the user must free these arrays
2854: once the matrix is destroyed
2856: You cannot set new nonzero locations into this matrix, that will generate an error.
2858: The i and j indices are 0 based
2860: .seealso: MatCreate(), MatCreateMPIAIJ(), MatCreateSeqAIJ()
2862: @*/
2863: int MatCreateSeqAIJWithArrays(MPI_Comm comm,int m,int n,int* i,int*j,PetscScalar *a,Mat *mat)
2864: {
2865: int ierr,ii;
2866: Mat_SeqAIJ *aij;
2869: MatCreate(comm,m,n,m,n,mat);
2870: MatSetType(*mat,MATSEQAIJ);
2871: MatSeqAIJSetPreallocation(*mat,SKIP_ALLOCATION,0);
2872: aij = (Mat_SeqAIJ*)(*mat)->data;
2874: if (i[0] != 0) {
2875: SETERRQ(1,"i (row indices) must start with 0");
2876: }
2877: aij->i = i;
2878: aij->j = j;
2879: aij->a = a;
2880: aij->singlemalloc = PETSC_FALSE;
2881: aij->nonew = -1; /*this indicates that inserting a new value in the matrix that generates a new nonzero is an error*/
2882: aij->freedata = PETSC_FALSE;
2884: for (ii=0; ii<m; ii++) {
2885: aij->ilen[ii] = aij->imax[ii] = i[ii+1] - i[ii];
2886: #if defined(PETSC_USE_BOPT_g)
2887: if (i[ii+1] - i[ii] < 0) SETERRQ2(1,"Negative row length in i (row indices) row = %d length = %d",ii,i[ii+1] - i[ii]);
2888: #endif
2889: }
2890: #if defined(PETSC_USE_BOPT_g)
2891: for (ii=0; ii<aij->i[m]; ii++) {
2892: if (j[ii] < 0) SETERRQ2(1,"Negative column index at location = %d index = %d",ii,j[ii]);
2893: if (j[ii] > n - 1) SETERRQ2(1,"Column index to large at location = %d index = %d",ii,j[ii]);
2894: }
2895: #endif
2897: MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);
2898: MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);
2899: return(0);
2900: }
2904: int MatSetColoring_SeqAIJ(Mat A,ISColoring coloring)
2905: {
2906: int ierr;
2907: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2910: if (coloring->ctype == IS_COLORING_LOCAL) {
2911: ISColoringReference(coloring);
2912: a->coloring = coloring;
2913: } else if (coloring->ctype == IS_COLORING_GHOSTED) {
2914: int i,*larray;
2915: ISColoring ocoloring;
2916: ISColoringValue *colors;
2918: /* set coloring for diagonal portion */
2919: PetscMalloc((A->n+1)*sizeof(int),&larray);
2920: for (i=0; i<A->n; i++) {
2921: larray[i] = i;
2922: }
2923: ISGlobalToLocalMappingApply(A->mapping,IS_GTOLM_MASK,A->n,larray,PETSC_NULL,larray);
2924: PetscMalloc((A->n+1)*sizeof(ISColoringValue),&colors);
2925: for (i=0; i<A->n; i++) {
2926: colors[i] = coloring->colors[larray[i]];
2927: }
2928: PetscFree(larray);
2929: ISColoringCreate(PETSC_COMM_SELF,A->n,colors,&ocoloring);
2930: a->coloring = ocoloring;
2931: }
2932: return(0);
2933: }
2935: #if defined(PETSC_HAVE_ADIC) && !defined(PETSC_USE_COMPLEX) && !defined(PETSC_USE_SINGLE)
2936: EXTERN_C_BEGIN
2937: #include "adic/ad_utils.h"
2938: EXTERN_C_END
2942: int MatSetValuesAdic_SeqAIJ(Mat A,void *advalues)
2943: {
2944: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2945: int m = A->m,*ii = a->i,*jj = a->j,nz,i,j,nlen;
2946: PetscScalar *v = a->a,*values = ((PetscScalar*)advalues)+1;
2947: ISColoringValue *color;
2950: if (!a->coloring) SETERRQ(1,"Coloring not set for matrix");
2951: nlen = PetscADGetDerivTypeSize()/sizeof(PetscScalar);
2952: color = a->coloring->colors;
2953: /* loop over rows */
2954: for (i=0; i<m; i++) {
2955: nz = ii[i+1] - ii[i];
2956: /* loop over columns putting computed value into matrix */
2957: for (j=0; j<nz; j++) {
2958: *v++ = values[color[*jj++]];
2959: }
2960: values += nlen; /* jump to next row of derivatives */
2961: }
2962: return(0);
2963: }
2965: #else
2969: int MatSetValuesAdic_SeqAIJ(Mat A,void *advalues)
2970: {
2972: SETERRQ(1,"PETSc installed without ADIC");
2973: }
2975: #endif
2979: int MatSetValuesAdifor_SeqAIJ(Mat A,int nl,void *advalues)
2980: {
2981: Mat_SeqAIJ *a = (Mat_SeqAIJ*)A->data;
2982: int m = A->m,*ii = a->i,*jj = a->j,nz,i,j;
2983: PetscScalar *v = a->a,*values = (PetscScalar *)advalues;
2984: ISColoringValue *color;
2987: if (!a->coloring) SETERRQ(1,"Coloring not set for matrix");
2988: color = a->coloring->colors;
2989: /* loop over rows */
2990: for (i=0; i<m; i++) {
2991: nz = ii[i+1] - ii[i];
2992: /* loop over columns putting computed value into matrix */
2993: for (j=0; j<nz; j++) {
2994: *v++ = values[color[*jj++]];
2995: }
2996: values += nl; /* jump to next row of derivatives */
2997: }
2998: return(0);
2999: }