Actual source code: baij.c
petsc-3.10.5 2019-03-28
2: /*
3: Defines the basic matrix operations for the BAIJ (compressed row)
4: matrix storage format.
5: */
6: #include <../src/mat/impls/baij/seq/baij.h>
7: #include <petscblaslapack.h>
8: #include <petsc/private/kernels/blockinvert.h>
9: #include <petsc/private/kernels/blockmatmult.h>
11: #if defined(PETSC_HAVE_HYPRE)
12: PETSC_INTERN PetscErrorCode MatConvert_AIJ_HYPRE(Mat,MatType,MatReuse,Mat*);
13: #endif
15: #if defined(PETSC_HAVE_MKL_SPARSE_OPTIMIZE)
16: PETSC_INTERN PetscErrorCode MatConvert_SeqBAIJ_SeqBAIJMKL(Mat,MatType,MatReuse,Mat*);
17: #endif
18: PETSC_INTERN PetscErrorCode MatConvert_XAIJ_IS(Mat,MatType,MatReuse,Mat*);
19: PETSC_INTERN PetscErrorCode MatPtAP_IS_XAIJ(Mat,Mat,MatReuse,PetscReal,Mat*);
21: PetscErrorCode MatInvertBlockDiagonal_SeqBAIJ(Mat A,const PetscScalar **values)
22: {
23: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*) A->data;
25: PetscInt *diag_offset,i,bs = A->rmap->bs,mbs = a->mbs,ipvt[5],bs2 = bs*bs,*v_pivots;
26: MatScalar *v = a->a,*odiag,*diag,work[25],*v_work;
27: PetscReal shift = 0.0;
28: PetscBool allowzeropivot,zeropivotdetected=PETSC_FALSE;
31: allowzeropivot = PetscNot(A->erroriffailure);
33: if (a->idiagvalid) {
34: if (values) *values = a->idiag;
35: return(0);
36: }
37: MatMarkDiagonal_SeqBAIJ(A);
38: diag_offset = a->diag;
39: if (!a->idiag) {
40: PetscMalloc1(bs2*mbs,&a->idiag);
41: PetscLogObjectMemory((PetscObject)A,bs2*mbs*sizeof(PetscScalar));
42: }
43: diag = a->idiag;
44: if (values) *values = a->idiag;
45: /* factor and invert each block */
46: switch (bs) {
47: case 1:
48: for (i=0; i<mbs; i++) {
49: odiag = v + 1*diag_offset[i];
50: diag[0] = odiag[0];
52: if (PetscAbsScalar(diag[0] + shift) < PETSC_MACHINE_EPSILON) {
53: if (allowzeropivot) {
54: A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
55: A->factorerror_zeropivot_value = PetscAbsScalar(diag[0]);
56: A->factorerror_zeropivot_row = i;
57: PetscInfo1(A,"Zero pivot, row %D\n",i);
58: } else SETERRQ3(PETSC_COMM_SELF,PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot, row %D pivot value %g tolerance %g",i,(double)PetscAbsScalar(diag[0]),(double)PETSC_MACHINE_EPSILON);
59: }
61: diag[0] = (PetscScalar)1.0 / (diag[0] + shift);
62: diag += 1;
63: }
64: break;
65: case 2:
66: for (i=0; i<mbs; i++) {
67: odiag = v + 4*diag_offset[i];
68: diag[0] = odiag[0]; diag[1] = odiag[1]; diag[2] = odiag[2]; diag[3] = odiag[3];
69: PetscKernel_A_gets_inverse_A_2(diag,shift,allowzeropivot,&zeropivotdetected);
70: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
71: diag += 4;
72: }
73: break;
74: case 3:
75: for (i=0; i<mbs; i++) {
76: odiag = v + 9*diag_offset[i];
77: diag[0] = odiag[0]; diag[1] = odiag[1]; diag[2] = odiag[2]; diag[3] = odiag[3];
78: diag[4] = odiag[4]; diag[5] = odiag[5]; diag[6] = odiag[6]; diag[7] = odiag[7];
79: diag[8] = odiag[8];
80: PetscKernel_A_gets_inverse_A_3(diag,shift,allowzeropivot,&zeropivotdetected);
81: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
82: diag += 9;
83: }
84: break;
85: case 4:
86: for (i=0; i<mbs; i++) {
87: odiag = v + 16*diag_offset[i];
88: PetscMemcpy(diag,odiag,16*sizeof(PetscScalar));
89: PetscKernel_A_gets_inverse_A_4(diag,shift,allowzeropivot,&zeropivotdetected);
90: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
91: diag += 16;
92: }
93: break;
94: case 5:
95: for (i=0; i<mbs; i++) {
96: odiag = v + 25*diag_offset[i];
97: PetscMemcpy(diag,odiag,25*sizeof(PetscScalar));
98: PetscKernel_A_gets_inverse_A_5(diag,ipvt,work,shift,allowzeropivot,&zeropivotdetected);
99: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
100: diag += 25;
101: }
102: break;
103: case 6:
104: for (i=0; i<mbs; i++) {
105: odiag = v + 36*diag_offset[i];
106: PetscMemcpy(diag,odiag,36*sizeof(PetscScalar));
107: PetscKernel_A_gets_inverse_A_6(diag,shift,allowzeropivot,&zeropivotdetected);
108: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
109: diag += 36;
110: }
111: break;
112: case 7:
113: for (i=0; i<mbs; i++) {
114: odiag = v + 49*diag_offset[i];
115: PetscMemcpy(diag,odiag,49*sizeof(PetscScalar));
116: PetscKernel_A_gets_inverse_A_7(diag,shift,allowzeropivot,&zeropivotdetected);
117: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
118: diag += 49;
119: }
120: break;
121: default:
122: PetscMalloc2(bs,&v_work,bs,&v_pivots);
123: for (i=0; i<mbs; i++) {
124: odiag = v + bs2*diag_offset[i];
125: PetscMemcpy(diag,odiag,bs2*sizeof(PetscScalar));
126: PetscKernel_A_gets_inverse_A(bs,diag,v_pivots,v_work,allowzeropivot,&zeropivotdetected);
127: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
128: diag += bs2;
129: }
130: PetscFree2(v_work,v_pivots);
131: }
132: a->idiagvalid = PETSC_TRUE;
133: return(0);
134: }
136: PetscErrorCode MatSOR_SeqBAIJ(Mat A,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,PetscInt its,PetscInt lits,Vec xx)
137: {
138: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
139: PetscScalar *x,*work,*w,*workt,*t;
140: const MatScalar *v,*aa = a->a, *idiag;
141: const PetscScalar *b,*xb;
142: PetscScalar s[7], xw[7]={0}; /* avoid some compilers thinking xw is uninitialized */
143: PetscErrorCode ierr;
144: PetscInt m = a->mbs,i,i2,nz,bs = A->rmap->bs,bs2 = bs*bs,k,j,idx,it;
145: const PetscInt *diag,*ai = a->i,*aj = a->j,*vi;
148: its = its*lits;
149: if (flag & SOR_EISENSTAT) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"No support yet for Eisenstat");
150: if (its <= 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Relaxation requires global its %D and local its %D both positive",its,lits);
151: if (fshift) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Sorry, no support for diagonal shift");
152: if (omega != 1.0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Sorry, no support for non-trivial relaxation factor");
153: if ((flag & SOR_APPLY_UPPER) || (flag & SOR_APPLY_LOWER)) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Sorry, no support for applying upper or lower triangular parts");
155: if (!a->idiagvalid) {MatInvertBlockDiagonal(A,NULL);}
157: if (!m) return(0);
158: diag = a->diag;
159: idiag = a->idiag;
160: k = PetscMax(A->rmap->n,A->cmap->n);
161: if (!a->mult_work) {
162: PetscMalloc1(k+1,&a->mult_work);
163: }
164: if (!a->sor_workt) {
165: PetscMalloc1(k,&a->sor_workt);
166: }
167: if (!a->sor_work) {
168: PetscMalloc1(bs,&a->sor_work);
169: }
170: work = a->mult_work;
171: t = a->sor_workt;
172: w = a->sor_work;
174: VecGetArray(xx,&x);
175: VecGetArrayRead(bb,&b);
177: if (flag & SOR_ZERO_INITIAL_GUESS) {
178: if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) {
179: switch (bs) {
180: case 1:
181: PetscKernel_v_gets_A_times_w_1(x,idiag,b);
182: t[0] = b[0];
183: i2 = 1;
184: idiag += 1;
185: for (i=1; i<m; i++) {
186: v = aa + ai[i];
187: vi = aj + ai[i];
188: nz = diag[i] - ai[i];
189: s[0] = b[i2];
190: for (j=0; j<nz; j++) {
191: xw[0] = x[vi[j]];
192: PetscKernel_v_gets_v_minus_A_times_w_1(s,(v+j),xw);
193: }
194: t[i2] = s[0];
195: PetscKernel_v_gets_A_times_w_1(xw,idiag,s);
196: x[i2] = xw[0];
197: idiag += 1;
198: i2 += 1;
199: }
200: break;
201: case 2:
202: PetscKernel_v_gets_A_times_w_2(x,idiag,b);
203: t[0] = b[0]; t[1] = b[1];
204: i2 = 2;
205: idiag += 4;
206: for (i=1; i<m; i++) {
207: v = aa + 4*ai[i];
208: vi = aj + ai[i];
209: nz = diag[i] - ai[i];
210: s[0] = b[i2]; s[1] = b[i2+1];
211: for (j=0; j<nz; j++) {
212: idx = 2*vi[j];
213: it = 4*j;
214: xw[0] = x[idx]; xw[1] = x[1+idx];
215: PetscKernel_v_gets_v_minus_A_times_w_2(s,(v+it),xw);
216: }
217: t[i2] = s[0]; t[i2+1] = s[1];
218: PetscKernel_v_gets_A_times_w_2(xw,idiag,s);
219: x[i2] = xw[0]; x[i2+1] = xw[1];
220: idiag += 4;
221: i2 += 2;
222: }
223: break;
224: case 3:
225: PetscKernel_v_gets_A_times_w_3(x,idiag,b);
226: t[0] = b[0]; t[1] = b[1]; t[2] = b[2];
227: i2 = 3;
228: idiag += 9;
229: for (i=1; i<m; i++) {
230: v = aa + 9*ai[i];
231: vi = aj + ai[i];
232: nz = diag[i] - ai[i];
233: s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2];
234: while (nz--) {
235: idx = 3*(*vi++);
236: xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx];
237: PetscKernel_v_gets_v_minus_A_times_w_3(s,v,xw);
238: v += 9;
239: }
240: t[i2] = s[0]; t[i2+1] = s[1]; t[i2+2] = s[2];
241: PetscKernel_v_gets_A_times_w_3(xw,idiag,s);
242: x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2];
243: idiag += 9;
244: i2 += 3;
245: }
246: break;
247: case 4:
248: PetscKernel_v_gets_A_times_w_4(x,idiag,b);
249: t[0] = b[0]; t[1] = b[1]; t[2] = b[2]; t[3] = b[3];
250: i2 = 4;
251: idiag += 16;
252: for (i=1; i<m; i++) {
253: v = aa + 16*ai[i];
254: vi = aj + ai[i];
255: nz = diag[i] - ai[i];
256: s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2]; s[3] = b[i2+3];
257: while (nz--) {
258: idx = 4*(*vi++);
259: xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx]; xw[3] = x[3+idx];
260: PetscKernel_v_gets_v_minus_A_times_w_4(s,v,xw);
261: v += 16;
262: }
263: t[i2] = s[0]; t[i2+1] = s[1]; t[i2+2] = s[2]; t[i2 + 3] = s[3];
264: PetscKernel_v_gets_A_times_w_4(xw,idiag,s);
265: x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2]; x[i2+3] = xw[3];
266: idiag += 16;
267: i2 += 4;
268: }
269: break;
270: case 5:
271: PetscKernel_v_gets_A_times_w_5(x,idiag,b);
272: t[0] = b[0]; t[1] = b[1]; t[2] = b[2]; t[3] = b[3]; t[4] = b[4];
273: i2 = 5;
274: idiag += 25;
275: for (i=1; i<m; i++) {
276: v = aa + 25*ai[i];
277: vi = aj + ai[i];
278: nz = diag[i] - ai[i];
279: s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2]; s[3] = b[i2+3]; s[4] = b[i2+4];
280: while (nz--) {
281: idx = 5*(*vi++);
282: xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx]; xw[3] = x[3+idx]; xw[4] = x[4+idx];
283: PetscKernel_v_gets_v_minus_A_times_w_5(s,v,xw);
284: v += 25;
285: }
286: t[i2] = s[0]; t[i2+1] = s[1]; t[i2+2] = s[2]; t[i2+3] = s[3]; t[i2+4] = s[4];
287: PetscKernel_v_gets_A_times_w_5(xw,idiag,s);
288: x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2]; x[i2+3] = xw[3]; x[i2+4] = xw[4];
289: idiag += 25;
290: i2 += 5;
291: }
292: break;
293: case 6:
294: PetscKernel_v_gets_A_times_w_6(x,idiag,b);
295: t[0] = b[0]; t[1] = b[1]; t[2] = b[2]; t[3] = b[3]; t[4] = b[4]; t[5] = b[5];
296: i2 = 6;
297: idiag += 36;
298: for (i=1; i<m; i++) {
299: v = aa + 36*ai[i];
300: vi = aj + ai[i];
301: nz = diag[i] - ai[i];
302: s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2]; s[3] = b[i2+3]; s[4] = b[i2+4]; s[5] = b[i2+5];
303: while (nz--) {
304: idx = 6*(*vi++);
305: xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx];
306: xw[3] = x[3+idx]; xw[4] = x[4+idx]; xw[5] = x[5+idx];
307: PetscKernel_v_gets_v_minus_A_times_w_6(s,v,xw);
308: v += 36;
309: }
310: t[i2] = s[0]; t[i2+1] = s[1]; t[i2+2] = s[2];
311: t[i2+3] = s[3]; t[i2+4] = s[4]; t[i2+5] = s[5];
312: PetscKernel_v_gets_A_times_w_6(xw,idiag,s);
313: x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2]; x[i2+3] = xw[3]; x[i2+4] = xw[4]; x[i2+5] = xw[5];
314: idiag += 36;
315: i2 += 6;
316: }
317: break;
318: case 7:
319: PetscKernel_v_gets_A_times_w_7(x,idiag,b);
320: t[0] = b[0]; t[1] = b[1]; t[2] = b[2];
321: t[3] = b[3]; t[4] = b[4]; t[5] = b[5]; t[6] = b[6];
322: i2 = 7;
323: idiag += 49;
324: for (i=1; i<m; i++) {
325: v = aa + 49*ai[i];
326: vi = aj + ai[i];
327: nz = diag[i] - ai[i];
328: s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2];
329: s[3] = b[i2+3]; s[4] = b[i2+4]; s[5] = b[i2+5]; s[6] = b[i2+6];
330: while (nz--) {
331: idx = 7*(*vi++);
332: xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx];
333: xw[3] = x[3+idx]; xw[4] = x[4+idx]; xw[5] = x[5+idx]; xw[6] = x[6+idx];
334: PetscKernel_v_gets_v_minus_A_times_w_7(s,v,xw);
335: v += 49;
336: }
337: t[i2] = s[0]; t[i2+1] = s[1]; t[i2+2] = s[2];
338: t[i2+3] = s[3]; t[i2+4] = s[4]; t[i2+5] = s[5]; t[i2+6] = s[6];
339: PetscKernel_v_gets_A_times_w_7(xw,idiag,s);
340: x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2];
341: x[i2+3] = xw[3]; x[i2+4] = xw[4]; x[i2+5] = xw[5]; x[i2+6] = xw[6];
342: idiag += 49;
343: i2 += 7;
344: }
345: break;
346: default:
347: PetscKernel_w_gets_Ar_times_v(bs,bs,b,idiag,x);
348: PetscMemcpy(t,b,bs*sizeof(PetscScalar));
349: i2 = bs;
350: idiag += bs2;
351: for (i=1; i<m; i++) {
352: v = aa + bs2*ai[i];
353: vi = aj + ai[i];
354: nz = diag[i] - ai[i];
356: PetscMemcpy(w,b+i2,bs*sizeof(PetscScalar));
357: /* copy all rows of x that are needed into contiguous space */
358: workt = work;
359: for (j=0; j<nz; j++) {
360: PetscMemcpy(workt,x + bs*(*vi++),bs*sizeof(PetscScalar));
361: workt += bs;
362: }
363: PetscKernel_w_gets_w_minus_Ar_times_v(bs,bs*nz,w,v,work);
364: PetscMemcpy(t+i2,w,bs*sizeof(PetscScalar));
365: PetscKernel_w_gets_Ar_times_v(bs,bs,w,idiag,x+i2);
367: idiag += bs2;
368: i2 += bs;
369: }
370: break;
371: }
372: /* for logging purposes assume number of nonzero in lower half is 1/2 of total */
373: PetscLogFlops(1.0*bs2*a->nz);
374: xb = t;
375: }
376: else xb = b;
377: if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) {
378: idiag = a->idiag+bs2*(a->mbs-1);
379: i2 = bs * (m-1);
380: switch (bs) {
381: case 1:
382: s[0] = xb[i2];
383: PetscKernel_v_gets_A_times_w_1(xw,idiag,s);
384: x[i2] = xw[0];
385: i2 -= 1;
386: for (i=m-2; i>=0; i--) {
387: v = aa + (diag[i]+1);
388: vi = aj + diag[i] + 1;
389: nz = ai[i+1] - diag[i] - 1;
390: s[0] = xb[i2];
391: for (j=0; j<nz; j++) {
392: xw[0] = x[vi[j]];
393: PetscKernel_v_gets_v_minus_A_times_w_1(s,(v+j),xw);
394: }
395: PetscKernel_v_gets_A_times_w_1(xw,idiag,s);
396: x[i2] = xw[0];
397: idiag -= 1;
398: i2 -= 1;
399: }
400: break;
401: case 2:
402: s[0] = xb[i2]; s[1] = xb[i2+1];
403: PetscKernel_v_gets_A_times_w_2(xw,idiag,s);
404: x[i2] = xw[0]; x[i2+1] = xw[1];
405: i2 -= 2;
406: idiag -= 4;
407: for (i=m-2; i>=0; i--) {
408: v = aa + 4*(diag[i] + 1);
409: vi = aj + diag[i] + 1;
410: nz = ai[i+1] - diag[i] - 1;
411: s[0] = xb[i2]; s[1] = xb[i2+1];
412: for (j=0; j<nz; j++) {
413: idx = 2*vi[j];
414: it = 4*j;
415: xw[0] = x[idx]; xw[1] = x[1+idx];
416: PetscKernel_v_gets_v_minus_A_times_w_2(s,(v+it),xw);
417: }
418: PetscKernel_v_gets_A_times_w_2(xw,idiag,s);
419: x[i2] = xw[0]; x[i2+1] = xw[1];
420: idiag -= 4;
421: i2 -= 2;
422: }
423: break;
424: case 3:
425: s[0] = xb[i2]; s[1] = xb[i2+1]; s[2] = xb[i2+2];
426: PetscKernel_v_gets_A_times_w_3(xw,idiag,s);
427: x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2];
428: i2 -= 3;
429: idiag -= 9;
430: for (i=m-2; i>=0; i--) {
431: v = aa + 9*(diag[i]+1);
432: vi = aj + diag[i] + 1;
433: nz = ai[i+1] - diag[i] - 1;
434: s[0] = xb[i2]; s[1] = xb[i2+1]; s[2] = xb[i2+2];
435: while (nz--) {
436: idx = 3*(*vi++);
437: xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx];
438: PetscKernel_v_gets_v_minus_A_times_w_3(s,v,xw);
439: v += 9;
440: }
441: PetscKernel_v_gets_A_times_w_3(xw,idiag,s);
442: x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2];
443: idiag -= 9;
444: i2 -= 3;
445: }
446: break;
447: case 4:
448: s[0] = xb[i2]; s[1] = xb[i2+1]; s[2] = xb[i2+2]; s[3] = xb[i2+3];
449: PetscKernel_v_gets_A_times_w_4(xw,idiag,s);
450: x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2]; x[i2+3] = xw[3];
451: i2 -= 4;
452: idiag -= 16;
453: for (i=m-2; i>=0; i--) {
454: v = aa + 16*(diag[i]+1);
455: vi = aj + diag[i] + 1;
456: nz = ai[i+1] - diag[i] - 1;
457: s[0] = xb[i2]; s[1] = xb[i2+1]; s[2] = xb[i2+2]; s[3] = xb[i2+3];
458: while (nz--) {
459: idx = 4*(*vi++);
460: xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx]; xw[3] = x[3+idx];
461: PetscKernel_v_gets_v_minus_A_times_w_4(s,v,xw);
462: v += 16;
463: }
464: PetscKernel_v_gets_A_times_w_4(xw,idiag,s);
465: x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2]; x[i2+3] = xw[3];
466: idiag -= 16;
467: i2 -= 4;
468: }
469: break;
470: case 5:
471: s[0] = xb[i2]; s[1] = xb[i2+1]; s[2] = xb[i2+2]; s[3] = xb[i2+3]; s[4] = xb[i2+4];
472: PetscKernel_v_gets_A_times_w_5(xw,idiag,s);
473: x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2]; x[i2+3] = xw[3]; x[i2+4] = xw[4];
474: i2 -= 5;
475: idiag -= 25;
476: for (i=m-2; i>=0; i--) {
477: v = aa + 25*(diag[i]+1);
478: vi = aj + diag[i] + 1;
479: nz = ai[i+1] - diag[i] - 1;
480: s[0] = xb[i2]; s[1] = xb[i2+1]; s[2] = xb[i2+2]; s[3] = xb[i2+3]; s[4] = xb[i2+4];
481: while (nz--) {
482: idx = 5*(*vi++);
483: xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx]; xw[3] = x[3+idx]; xw[4] = x[4+idx];
484: PetscKernel_v_gets_v_minus_A_times_w_5(s,v,xw);
485: v += 25;
486: }
487: PetscKernel_v_gets_A_times_w_5(xw,idiag,s);
488: x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2]; x[i2+3] = xw[3]; x[i2+4] = xw[4];
489: idiag -= 25;
490: i2 -= 5;
491: }
492: break;
493: case 6:
494: s[0] = xb[i2]; s[1] = xb[i2+1]; s[2] = xb[i2+2]; s[3] = xb[i2+3]; s[4] = xb[i2+4]; s[5] = xb[i2+5];
495: PetscKernel_v_gets_A_times_w_6(xw,idiag,s);
496: x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2]; x[i2+3] = xw[3]; x[i2+4] = xw[4]; x[i2+5] = xw[5];
497: i2 -= 6;
498: idiag -= 36;
499: for (i=m-2; i>=0; i--) {
500: v = aa + 36*(diag[i]+1);
501: vi = aj + diag[i] + 1;
502: nz = ai[i+1] - diag[i] - 1;
503: s[0] = xb[i2]; s[1] = xb[i2+1]; s[2] = xb[i2+2]; s[3] = xb[i2+3]; s[4] = xb[i2+4]; s[5] = xb[i2+5];
504: while (nz--) {
505: idx = 6*(*vi++);
506: xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx];
507: xw[3] = x[3+idx]; xw[4] = x[4+idx]; xw[5] = x[5+idx];
508: PetscKernel_v_gets_v_minus_A_times_w_6(s,v,xw);
509: v += 36;
510: }
511: PetscKernel_v_gets_A_times_w_6(xw,idiag,s);
512: x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2]; x[i2+3] = xw[3]; x[i2+4] = xw[4]; x[i2+5] = xw[5];
513: idiag -= 36;
514: i2 -= 6;
515: }
516: break;
517: case 7:
518: s[0] = xb[i2]; s[1] = xb[i2+1]; s[2] = xb[i2+2];
519: s[3] = xb[i2+3]; s[4] = xb[i2+4]; s[5] = xb[i2+5]; s[6] = xb[i2+6];
520: PetscKernel_v_gets_A_times_w_7(x,idiag,b);
521: x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2];
522: x[i2+3] = xw[3]; x[i2+4] = xw[4]; x[i2+5] = xw[5]; x[i2+6] = xw[6];
523: i2 -= 7;
524: idiag -= 49;
525: for (i=m-2; i>=0; i--) {
526: v = aa + 49*(diag[i]+1);
527: vi = aj + diag[i] + 1;
528: nz = ai[i+1] - diag[i] - 1;
529: s[0] = xb[i2]; s[1] = xb[i2+1]; s[2] = xb[i2+2];
530: s[3] = xb[i2+3]; s[4] = xb[i2+4]; s[5] = xb[i2+5]; s[6] = xb[i2+6];
531: while (nz--) {
532: idx = 7*(*vi++);
533: xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx];
534: xw[3] = x[3+idx]; xw[4] = x[4+idx]; xw[5] = x[5+idx]; xw[6] = x[6+idx];
535: PetscKernel_v_gets_v_minus_A_times_w_7(s,v,xw);
536: v += 49;
537: }
538: PetscKernel_v_gets_A_times_w_7(xw,idiag,s);
539: x[i2] = xw[0]; x[i2+1] = xw[1]; x[i2+2] = xw[2];
540: x[i2+3] = xw[3]; x[i2+4] = xw[4]; x[i2+5] = xw[5]; x[i2+6] = xw[6];
541: idiag -= 49;
542: i2 -= 7;
543: }
544: break;
545: default:
546: PetscMemcpy(w,xb+i2,bs*sizeof(PetscScalar));
547: PetscKernel_w_gets_Ar_times_v(bs,bs,w,idiag,x+i2);
548: i2 -= bs;
549: idiag -= bs2;
550: for (i=m-2; i>=0; i--) {
551: v = aa + bs2*(diag[i]+1);
552: vi = aj + diag[i] + 1;
553: nz = ai[i+1] - diag[i] - 1;
555: PetscMemcpy(w,xb+i2,bs*sizeof(PetscScalar));
556: /* copy all rows of x that are needed into contiguous space */
557: workt = work;
558: for (j=0; j<nz; j++) {
559: PetscMemcpy(workt,x + bs*(*vi++),bs*sizeof(PetscScalar));
560: workt += bs;
561: }
562: PetscKernel_w_gets_w_minus_Ar_times_v(bs,bs*nz,w,v,work);
563: PetscKernel_w_gets_Ar_times_v(bs,bs,w,idiag,x+i2);
565: idiag -= bs2;
566: i2 -= bs;
567: }
568: break;
569: }
570: PetscLogFlops(1.0*bs2*(a->nz));
571: }
572: its--;
573: }
574: while (its--) {
575: if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) {
576: idiag = a->idiag;
577: i2 = 0;
578: switch (bs) {
579: case 1:
580: for (i=0; i<m; i++) {
581: v = aa + ai[i];
582: vi = aj + ai[i];
583: nz = ai[i+1] - ai[i];
584: s[0] = b[i2];
585: for (j=0; j<nz; j++) {
586: xw[0] = x[vi[j]];
587: PetscKernel_v_gets_v_minus_A_times_w_1(s,(v+j),xw);
588: }
589: PetscKernel_v_gets_A_times_w_1(xw,idiag,s);
590: x[i2] += xw[0];
591: idiag += 1;
592: i2 += 1;
593: }
594: break;
595: case 2:
596: for (i=0; i<m; i++) {
597: v = aa + 4*ai[i];
598: vi = aj + ai[i];
599: nz = ai[i+1] - ai[i];
600: s[0] = b[i2]; s[1] = b[i2+1];
601: for (j=0; j<nz; j++) {
602: idx = 2*vi[j];
603: it = 4*j;
604: xw[0] = x[idx]; xw[1] = x[1+idx];
605: PetscKernel_v_gets_v_minus_A_times_w_2(s,(v+it),xw);
606: }
607: PetscKernel_v_gets_A_times_w_2(xw,idiag,s);
608: x[i2] += xw[0]; x[i2+1] += xw[1];
609: idiag += 4;
610: i2 += 2;
611: }
612: break;
613: case 3:
614: for (i=0; i<m; i++) {
615: v = aa + 9*ai[i];
616: vi = aj + ai[i];
617: nz = ai[i+1] - ai[i];
618: s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2];
619: while (nz--) {
620: idx = 3*(*vi++);
621: xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx];
622: PetscKernel_v_gets_v_minus_A_times_w_3(s,v,xw);
623: v += 9;
624: }
625: PetscKernel_v_gets_A_times_w_3(xw,idiag,s);
626: x[i2] += xw[0]; x[i2+1] += xw[1]; x[i2+2] += xw[2];
627: idiag += 9;
628: i2 += 3;
629: }
630: break;
631: case 4:
632: for (i=0; i<m; i++) {
633: v = aa + 16*ai[i];
634: vi = aj + ai[i];
635: nz = ai[i+1] - ai[i];
636: s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2]; s[3] = b[i2+3];
637: while (nz--) {
638: idx = 4*(*vi++);
639: xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx]; xw[3] = x[3+idx];
640: PetscKernel_v_gets_v_minus_A_times_w_4(s,v,xw);
641: v += 16;
642: }
643: PetscKernel_v_gets_A_times_w_4(xw,idiag,s);
644: x[i2] += xw[0]; x[i2+1] += xw[1]; x[i2+2] += xw[2]; x[i2+3] += xw[3];
645: idiag += 16;
646: i2 += 4;
647: }
648: break;
649: case 5:
650: for (i=0; i<m; i++) {
651: v = aa + 25*ai[i];
652: vi = aj + ai[i];
653: nz = ai[i+1] - ai[i];
654: s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2]; s[3] = b[i2+3]; s[4] = b[i2+4];
655: while (nz--) {
656: idx = 5*(*vi++);
657: xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx]; xw[3] = x[3+idx]; xw[4] = x[4+idx];
658: PetscKernel_v_gets_v_minus_A_times_w_5(s,v,xw);
659: v += 25;
660: }
661: PetscKernel_v_gets_A_times_w_5(xw,idiag,s);
662: x[i2] += xw[0]; x[i2+1] += xw[1]; x[i2+2] += xw[2]; x[i2+3] += xw[3]; x[i2+4] += xw[4];
663: idiag += 25;
664: i2 += 5;
665: }
666: break;
667: case 6:
668: for (i=0; i<m; i++) {
669: v = aa + 36*ai[i];
670: vi = aj + ai[i];
671: nz = ai[i+1] - ai[i];
672: s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2]; s[3] = b[i2+3]; s[4] = b[i2+4]; s[5] = b[i2+5];
673: while (nz--) {
674: idx = 6*(*vi++);
675: xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx];
676: xw[3] = x[3+idx]; xw[4] = x[4+idx]; xw[5] = x[5+idx];
677: PetscKernel_v_gets_v_minus_A_times_w_6(s,v,xw);
678: v += 36;
679: }
680: PetscKernel_v_gets_A_times_w_6(xw,idiag,s);
681: x[i2] += xw[0]; x[i2+1] += xw[1]; x[i2+2] += xw[2];
682: x[i2+3] += xw[3]; x[i2+4] += xw[4]; x[i2+5] += xw[5];
683: idiag += 36;
684: i2 += 6;
685: }
686: break;
687: case 7:
688: for (i=0; i<m; i++) {
689: v = aa + 49*ai[i];
690: vi = aj + ai[i];
691: nz = ai[i+1] - ai[i];
692: s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2];
693: s[3] = b[i2+3]; s[4] = b[i2+4]; s[5] = b[i2+5]; s[6] = b[i2+6];
694: while (nz--) {
695: idx = 7*(*vi++);
696: xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx];
697: xw[3] = x[3+idx]; xw[4] = x[4+idx]; xw[5] = x[5+idx]; xw[6] = x[6+idx];
698: PetscKernel_v_gets_v_minus_A_times_w_7(s,v,xw);
699: v += 49;
700: }
701: PetscKernel_v_gets_A_times_w_7(xw,idiag,s);
702: x[i2] += xw[0]; x[i2+1] += xw[1]; x[i2+2] += xw[2];
703: x[i2+3] += xw[3]; x[i2+4] += xw[4]; x[i2+5] += xw[5]; x[i2+6] += xw[6];
704: idiag += 49;
705: i2 += 7;
706: }
707: break;
708: default:
709: for (i=0; i<m; i++) {
710: v = aa + bs2*ai[i];
711: vi = aj + ai[i];
712: nz = ai[i+1] - ai[i];
714: PetscMemcpy(w,b+i2,bs*sizeof(PetscScalar));
715: /* copy all rows of x that are needed into contiguous space */
716: workt = work;
717: for (j=0; j<nz; j++) {
718: PetscMemcpy(workt,x + bs*(*vi++),bs*sizeof(PetscScalar));
719: workt += bs;
720: }
721: PetscKernel_w_gets_w_minus_Ar_times_v(bs,bs*nz,w,v,work);
722: PetscKernel_w_gets_w_plus_Ar_times_v(bs,bs,w,idiag,x+i2);
724: idiag += bs2;
725: i2 += bs;
726: }
727: break;
728: }
729: PetscLogFlops(2.0*bs2*a->nz);
730: }
731: if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) {
732: idiag = a->idiag+bs2*(a->mbs-1);
733: i2 = bs * (m-1);
734: switch (bs) {
735: case 1:
736: for (i=m-1; i>=0; i--) {
737: v = aa + ai[i];
738: vi = aj + ai[i];
739: nz = ai[i+1] - ai[i];
740: s[0] = b[i2];
741: for (j=0; j<nz; j++) {
742: xw[0] = x[vi[j]];
743: PetscKernel_v_gets_v_minus_A_times_w_1(s,(v+j),xw);
744: }
745: PetscKernel_v_gets_A_times_w_1(xw,idiag,s);
746: x[i2] += xw[0];
747: idiag -= 1;
748: i2 -= 1;
749: }
750: break;
751: case 2:
752: for (i=m-1; i>=0; i--) {
753: v = aa + 4*ai[i];
754: vi = aj + ai[i];
755: nz = ai[i+1] - ai[i];
756: s[0] = b[i2]; s[1] = b[i2+1];
757: for (j=0; j<nz; j++) {
758: idx = 2*vi[j];
759: it = 4*j;
760: xw[0] = x[idx]; xw[1] = x[1+idx];
761: PetscKernel_v_gets_v_minus_A_times_w_2(s,(v+it),xw);
762: }
763: PetscKernel_v_gets_A_times_w_2(xw,idiag,s);
764: x[i2] += xw[0]; x[i2+1] += xw[1];
765: idiag -= 4;
766: i2 -= 2;
767: }
768: break;
769: case 3:
770: for (i=m-1; i>=0; i--) {
771: v = aa + 9*ai[i];
772: vi = aj + ai[i];
773: nz = ai[i+1] - ai[i];
774: s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2];
775: while (nz--) {
776: idx = 3*(*vi++);
777: xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx];
778: PetscKernel_v_gets_v_minus_A_times_w_3(s,v,xw);
779: v += 9;
780: }
781: PetscKernel_v_gets_A_times_w_3(xw,idiag,s);
782: x[i2] += xw[0]; x[i2+1] += xw[1]; x[i2+2] += xw[2];
783: idiag -= 9;
784: i2 -= 3;
785: }
786: break;
787: case 4:
788: for (i=m-1; i>=0; i--) {
789: v = aa + 16*ai[i];
790: vi = aj + ai[i];
791: nz = ai[i+1] - ai[i];
792: s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2]; s[3] = b[i2+3];
793: while (nz--) {
794: idx = 4*(*vi++);
795: xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx]; xw[3] = x[3+idx];
796: PetscKernel_v_gets_v_minus_A_times_w_4(s,v,xw);
797: v += 16;
798: }
799: PetscKernel_v_gets_A_times_w_4(xw,idiag,s);
800: x[i2] += xw[0]; x[i2+1] += xw[1]; x[i2+2] += xw[2]; x[i2+3] += xw[3];
801: idiag -= 16;
802: i2 -= 4;
803: }
804: break;
805: case 5:
806: for (i=m-1; i>=0; i--) {
807: v = aa + 25*ai[i];
808: vi = aj + ai[i];
809: nz = ai[i+1] - ai[i];
810: s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2]; s[3] = b[i2+3]; s[4] = b[i2+4];
811: while (nz--) {
812: idx = 5*(*vi++);
813: xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx]; xw[3] = x[3+idx]; xw[4] = x[4+idx];
814: PetscKernel_v_gets_v_minus_A_times_w_5(s,v,xw);
815: v += 25;
816: }
817: PetscKernel_v_gets_A_times_w_5(xw,idiag,s);
818: x[i2] += xw[0]; x[i2+1] += xw[1]; x[i2+2] += xw[2]; x[i2+3] += xw[3]; x[i2+4] += xw[4];
819: idiag -= 25;
820: i2 -= 5;
821: }
822: break;
823: case 6:
824: for (i=m-1; i>=0; i--) {
825: v = aa + 36*ai[i];
826: vi = aj + ai[i];
827: nz = ai[i+1] - ai[i];
828: s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2]; s[3] = b[i2+3]; s[4] = b[i2+4]; s[5] = b[i2+5];
829: while (nz--) {
830: idx = 6*(*vi++);
831: xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx];
832: xw[3] = x[3+idx]; xw[4] = x[4+idx]; xw[5] = x[5+idx];
833: PetscKernel_v_gets_v_minus_A_times_w_6(s,v,xw);
834: v += 36;
835: }
836: PetscKernel_v_gets_A_times_w_6(xw,idiag,s);
837: x[i2] += xw[0]; x[i2+1] += xw[1]; x[i2+2] += xw[2];
838: x[i2+3] += xw[3]; x[i2+4] += xw[4]; x[i2+5] += xw[5];
839: idiag -= 36;
840: i2 -= 6;
841: }
842: break;
843: case 7:
844: for (i=m-1; i>=0; i--) {
845: v = aa + 49*ai[i];
846: vi = aj + ai[i];
847: nz = ai[i+1] - ai[i];
848: s[0] = b[i2]; s[1] = b[i2+1]; s[2] = b[i2+2];
849: s[3] = b[i2+3]; s[4] = b[i2+4]; s[5] = b[i2+5]; s[6] = b[i2+6];
850: while (nz--) {
851: idx = 7*(*vi++);
852: xw[0] = x[idx]; xw[1] = x[1+idx]; xw[2] = x[2+idx];
853: xw[3] = x[3+idx]; xw[4] = x[4+idx]; xw[5] = x[5+idx]; xw[6] = x[6+idx];
854: PetscKernel_v_gets_v_minus_A_times_w_7(s,v,xw);
855: v += 49;
856: }
857: PetscKernel_v_gets_A_times_w_7(xw,idiag,s);
858: x[i2] += xw[0]; x[i2+1] += xw[1]; x[i2+2] += xw[2];
859: x[i2+3] += xw[3]; x[i2+4] += xw[4]; x[i2+5] += xw[5]; x[i2+6] += xw[6];
860: idiag -= 49;
861: i2 -= 7;
862: }
863: break;
864: default:
865: for (i=m-1; i>=0; i--) {
866: v = aa + bs2*ai[i];
867: vi = aj + ai[i];
868: nz = ai[i+1] - ai[i];
870: PetscMemcpy(w,b+i2,bs*sizeof(PetscScalar));
871: /* copy all rows of x that are needed into contiguous space */
872: workt = work;
873: for (j=0; j<nz; j++) {
874: PetscMemcpy(workt,x + bs*(*vi++),bs*sizeof(PetscScalar));
875: workt += bs;
876: }
877: PetscKernel_w_gets_w_minus_Ar_times_v(bs,bs*nz,w,v,work);
878: PetscKernel_w_gets_w_plus_Ar_times_v(bs,bs,w,idiag,x+i2);
880: idiag -= bs2;
881: i2 -= bs;
882: }
883: break;
884: }
885: PetscLogFlops(2.0*bs2*(a->nz));
886: }
887: }
888: VecRestoreArray(xx,&x);
889: VecRestoreArrayRead(bb,&b);
890: return(0);
891: }
894: /*
895: Special version for direct calls from Fortran (Used in PETSc-fun3d)
896: */
897: #if defined(PETSC_HAVE_FORTRAN_CAPS)
898: #define matsetvaluesblocked4_ MATSETVALUESBLOCKED4
899: #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE)
900: #define matsetvaluesblocked4_ matsetvaluesblocked4
901: #endif
903: PETSC_EXTERN void matsetvaluesblocked4_(Mat *AA,PetscInt *mm,const PetscInt im[],PetscInt *nn,const PetscInt in[],const PetscScalar v[])
904: {
905: Mat A = *AA;
906: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
907: PetscInt *rp,k,low,high,t,ii,jj,row,nrow,i,col,l,N,m = *mm,n = *nn;
908: PetscInt *ai =a->i,*ailen=a->ilen;
909: PetscInt *aj =a->j,stepval,lastcol = -1;
910: const PetscScalar *value = v;
911: MatScalar *ap,*aa = a->a,*bap;
914: if (A->rmap->bs != 4) SETERRABORT(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Can only be called with a block size of 4");
915: stepval = (n-1)*4;
916: for (k=0; k<m; k++) { /* loop over added rows */
917: row = im[k];
918: rp = aj + ai[row];
919: ap = aa + 16*ai[row];
920: nrow = ailen[row];
921: low = 0;
922: high = nrow;
923: for (l=0; l<n; l++) { /* loop over added columns */
924: col = in[l];
925: if (col <= lastcol) low = 0;
926: else high = nrow;
927: lastcol = col;
928: value = v + k*(stepval+4 + l)*4;
929: while (high-low > 7) {
930: t = (low+high)/2;
931: if (rp[t] > col) high = t;
932: else low = t;
933: }
934: for (i=low; i<high; i++) {
935: if (rp[i] > col) break;
936: if (rp[i] == col) {
937: bap = ap + 16*i;
938: for (ii=0; ii<4; ii++,value+=stepval) {
939: for (jj=ii; jj<16; jj+=4) {
940: bap[jj] += *value++;
941: }
942: }
943: goto noinsert2;
944: }
945: }
946: N = nrow++ - 1;
947: high++; /* added new column index thus must search to one higher than before */
948: /* shift up all the later entries in this row */
949: for (ii=N; ii>=i; ii--) {
950: rp[ii+1] = rp[ii];
951: PetscMemcpy(ap+16*(ii+1),ap+16*(ii),16*sizeof(MatScalar));
952: }
953: if (N >= i) {
954: PetscMemzero(ap+16*i,16*sizeof(MatScalar));
955: }
956: rp[i] = col;
957: bap = ap + 16*i;
958: for (ii=0; ii<4; ii++,value+=stepval) {
959: for (jj=ii; jj<16; jj+=4) {
960: bap[jj] = *value++;
961: }
962: }
963: noinsert2:;
964: low = i;
965: }
966: ailen[row] = nrow;
967: }
968: PetscFunctionReturnVoid();
969: }
971: #if defined(PETSC_HAVE_FORTRAN_CAPS)
972: #define matsetvalues4_ MATSETVALUES4
973: #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE)
974: #define matsetvalues4_ matsetvalues4
975: #endif
977: PETSC_EXTERN void matsetvalues4_(Mat *AA,PetscInt *mm,PetscInt *im,PetscInt *nn,PetscInt *in,PetscScalar *v)
978: {
979: Mat A = *AA;
980: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
981: PetscInt *rp,k,low,high,t,ii,row,nrow,i,col,l,N,n = *nn,m = *mm;
982: PetscInt *ai=a->i,*ailen=a->ilen;
983: PetscInt *aj=a->j,brow,bcol;
984: PetscInt ridx,cidx,lastcol = -1;
985: MatScalar *ap,value,*aa=a->a,*bap;
988: for (k=0; k<m; k++) { /* loop over added rows */
989: row = im[k]; brow = row/4;
990: rp = aj + ai[brow];
991: ap = aa + 16*ai[brow];
992: nrow = ailen[brow];
993: low = 0;
994: high = nrow;
995: for (l=0; l<n; l++) { /* loop over added columns */
996: col = in[l]; bcol = col/4;
997: ridx = row % 4; cidx = col % 4;
998: value = v[l + k*n];
999: if (col <= lastcol) low = 0;
1000: else high = nrow;
1001: lastcol = col;
1002: while (high-low > 7) {
1003: t = (low+high)/2;
1004: if (rp[t] > bcol) high = t;
1005: else low = t;
1006: }
1007: for (i=low; i<high; i++) {
1008: if (rp[i] > bcol) break;
1009: if (rp[i] == bcol) {
1010: bap = ap + 16*i + 4*cidx + ridx;
1011: *bap += value;
1012: goto noinsert1;
1013: }
1014: }
1015: N = nrow++ - 1;
1016: high++; /* added new column thus must search to one higher than before */
1017: /* shift up all the later entries in this row */
1018: for (ii=N; ii>=i; ii--) {
1019: rp[ii+1] = rp[ii];
1020: PetscMemcpy(ap+16*(ii+1),ap+16*(ii),16*sizeof(MatScalar));
1021: }
1022: if (N>=i) {
1023: PetscMemzero(ap+16*i,16*sizeof(MatScalar));
1024: }
1025: rp[i] = bcol;
1026: ap[16*i + 4*cidx + ridx] = value;
1027: noinsert1:;
1028: low = i;
1029: }
1030: ailen[brow] = nrow;
1031: }
1032: PetscFunctionReturnVoid();
1033: }
1035: /*
1036: Checks for missing diagonals
1037: */
1038: PetscErrorCode MatMissingDiagonal_SeqBAIJ(Mat A,PetscBool *missing,PetscInt *d)
1039: {
1040: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
1042: PetscInt *diag,*ii = a->i,i;
1045: MatMarkDiagonal_SeqBAIJ(A);
1046: *missing = PETSC_FALSE;
1047: if (A->rmap->n > 0 && !ii) {
1048: *missing = PETSC_TRUE;
1049: if (d) *d = 0;
1050: PetscInfo(A,"Matrix has no entries therefore is missing diagonal\n");
1051: } else {
1052: diag = a->diag;
1053: for (i=0; i<a->mbs; i++) {
1054: if (diag[i] >= ii[i+1]) {
1055: *missing = PETSC_TRUE;
1056: if (d) *d = i;
1057: PetscInfo1(A,"Matrix is missing block diagonal number %D\n",i);
1058: break;
1059: }
1060: }
1061: }
1062: return(0);
1063: }
1065: PetscErrorCode MatMarkDiagonal_SeqBAIJ(Mat A)
1066: {
1067: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
1069: PetscInt i,j,m = a->mbs;
1072: if (!a->diag) {
1073: PetscMalloc1(m,&a->diag);
1074: PetscLogObjectMemory((PetscObject)A,m*sizeof(PetscInt));
1075: a->free_diag = PETSC_TRUE;
1076: }
1077: for (i=0; i<m; i++) {
1078: a->diag[i] = a->i[i+1];
1079: for (j=a->i[i]; j<a->i[i+1]; j++) {
1080: if (a->j[j] == i) {
1081: a->diag[i] = j;
1082: break;
1083: }
1084: }
1085: }
1086: return(0);
1087: }
1090: static PetscErrorCode MatGetRowIJ_SeqBAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool blockcompressed,PetscInt *nn,const PetscInt *inia[],const PetscInt *inja[],PetscBool *done)
1091: {
1092: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
1094: PetscInt i,j,n = a->mbs,nz = a->i[n],*tia,*tja,bs = A->rmap->bs,k,l,cnt;
1095: PetscInt **ia = (PetscInt**)inia,**ja = (PetscInt**)inja;
1098: *nn = n;
1099: if (!ia) return(0);
1100: if (symmetric) {
1101: MatToSymmetricIJ_SeqAIJ(n,a->i,a->j,PETSC_TRUE,0,0,&tia,&tja);
1102: nz = tia[n];
1103: } else {
1104: tia = a->i; tja = a->j;
1105: }
1107: if (!blockcompressed && bs > 1) {
1108: (*nn) *= bs;
1109: /* malloc & create the natural set of indices */
1110: PetscMalloc1((n+1)*bs,ia);
1111: if (n) {
1112: (*ia)[0] = oshift;
1113: for (j=1; j<bs; j++) {
1114: (*ia)[j] = (tia[1]-tia[0])*bs+(*ia)[j-1];
1115: }
1116: }
1118: for (i=1; i<n; i++) {
1119: (*ia)[i*bs] = (tia[i]-tia[i-1])*bs + (*ia)[i*bs-1];
1120: for (j=1; j<bs; j++) {
1121: (*ia)[i*bs+j] = (tia[i+1]-tia[i])*bs + (*ia)[i*bs+j-1];
1122: }
1123: }
1124: if (n) {
1125: (*ia)[n*bs] = (tia[n]-tia[n-1])*bs + (*ia)[n*bs-1];
1126: }
1128: if (inja) {
1129: PetscMalloc1(nz*bs*bs,ja);
1130: cnt = 0;
1131: for (i=0; i<n; i++) {
1132: for (j=0; j<bs; j++) {
1133: for (k=tia[i]; k<tia[i+1]; k++) {
1134: for (l=0; l<bs; l++) {
1135: (*ja)[cnt++] = bs*tja[k] + l;
1136: }
1137: }
1138: }
1139: }
1140: }
1142: if (symmetric) { /* deallocate memory allocated in MatToSymmetricIJ_SeqAIJ() */
1143: PetscFree(tia);
1144: PetscFree(tja);
1145: }
1146: } else if (oshift == 1) {
1147: if (symmetric) {
1148: nz = tia[A->rmap->n/bs];
1149: /* add 1 to i and j indices */
1150: for (i=0; i<A->rmap->n/bs+1; i++) tia[i] = tia[i] + 1;
1151: *ia = tia;
1152: if (ja) {
1153: for (i=0; i<nz; i++) tja[i] = tja[i] + 1;
1154: *ja = tja;
1155: }
1156: } else {
1157: nz = a->i[A->rmap->n/bs];
1158: /* malloc space and add 1 to i and j indices */
1159: PetscMalloc1(A->rmap->n/bs+1,ia);
1160: for (i=0; i<A->rmap->n/bs+1; i++) (*ia)[i] = a->i[i] + 1;
1161: if (ja) {
1162: PetscMalloc1(nz,ja);
1163: for (i=0; i<nz; i++) (*ja)[i] = a->j[i] + 1;
1164: }
1165: }
1166: } else {
1167: *ia = tia;
1168: if (ja) *ja = tja;
1169: }
1170: return(0);
1171: }
1173: static PetscErrorCode MatRestoreRowIJ_SeqBAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool blockcompressed,PetscInt *nn,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done)
1174: {
1178: if (!ia) return(0);
1179: if ((!blockcompressed && A->rmap->bs > 1) || (symmetric || oshift == 1)) {
1180: PetscFree(*ia);
1181: if (ja) {PetscFree(*ja);}
1182: }
1183: return(0);
1184: }
1186: PetscErrorCode MatDestroy_SeqBAIJ(Mat A)
1187: {
1188: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
1192: #if defined(PETSC_USE_LOG)
1193: PetscLogObjectState((PetscObject)A,"Rows=%D, Cols=%D, NZ=%D",A->rmap->N,A->cmap->n,a->nz);
1194: #endif
1195: MatSeqXAIJFreeAIJ(A,&a->a,&a->j,&a->i);
1196: ISDestroy(&a->row);
1197: ISDestroy(&a->col);
1198: if (a->free_diag) {PetscFree(a->diag);}
1199: PetscFree(a->idiag);
1200: if (a->free_imax_ilen) {PetscFree2(a->imax,a->ilen);}
1201: PetscFree(a->solve_work);
1202: PetscFree(a->mult_work);
1203: PetscFree(a->sor_workt);
1204: PetscFree(a->sor_work);
1205: ISDestroy(&a->icol);
1206: PetscFree(a->saved_values);
1207: PetscFree2(a->compressedrow.i,a->compressedrow.rindex);
1209: MatDestroy(&a->sbaijMat);
1210: MatDestroy(&a->parent);
1211: PetscFree(A->data);
1213: PetscObjectChangeTypeName((PetscObject)A,0);
1214: PetscObjectComposeFunction((PetscObject)A,"MatInvertBlockDiagonal_C",NULL);
1215: PetscObjectComposeFunction((PetscObject)A,"MatStoreValues_C",NULL);
1216: PetscObjectComposeFunction((PetscObject)A,"MatRetrieveValues_C",NULL);
1217: PetscObjectComposeFunction((PetscObject)A,"MatSeqBAIJSetColumnIndices_C",NULL);
1218: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqbaij_seqaij_C",NULL);
1219: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqbaij_seqsbaij_C",NULL);
1220: PetscObjectComposeFunction((PetscObject)A,"MatSeqBAIJSetPreallocation_C",NULL);
1221: PetscObjectComposeFunction((PetscObject)A,"MatSeqBAIJSetPreallocationCSR_C",NULL);
1222: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqbaij_seqbstrm_C",NULL);
1223: PetscObjectComposeFunction((PetscObject)A,"MatIsTranspose_C",NULL);
1224: #if defined(PETSC_HAVE_HYPRE)
1225: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqbaij_hypre_C",NULL);
1226: #endif
1227: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqbaij_is_C",NULL);
1228: PetscObjectComposeFunction((PetscObject)A,"MatPtAP_is_seqaij_C",NULL);
1229: return(0);
1230: }
1232: PetscErrorCode MatSetOption_SeqBAIJ(Mat A,MatOption op,PetscBool flg)
1233: {
1234: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
1238: switch (op) {
1239: case MAT_ROW_ORIENTED:
1240: a->roworiented = flg;
1241: break;
1242: case MAT_KEEP_NONZERO_PATTERN:
1243: a->keepnonzeropattern = flg;
1244: break;
1245: case MAT_NEW_NONZERO_LOCATIONS:
1246: a->nonew = (flg ? 0 : 1);
1247: break;
1248: case MAT_NEW_NONZERO_LOCATION_ERR:
1249: a->nonew = (flg ? -1 : 0);
1250: break;
1251: case MAT_NEW_NONZERO_ALLOCATION_ERR:
1252: a->nonew = (flg ? -2 : 0);
1253: break;
1254: case MAT_UNUSED_NONZERO_LOCATION_ERR:
1255: a->nounused = (flg ? -1 : 0);
1256: break;
1257: case MAT_NEW_DIAGONALS:
1258: case MAT_IGNORE_OFF_PROC_ENTRIES:
1259: case MAT_USE_HASH_TABLE:
1260: PetscInfo1(A,"Option %s ignored\n",MatOptions[op]);
1261: break;
1262: case MAT_SPD:
1263: case MAT_SYMMETRIC:
1264: case MAT_STRUCTURALLY_SYMMETRIC:
1265: case MAT_HERMITIAN:
1266: case MAT_SYMMETRY_ETERNAL:
1267: case MAT_SUBMAT_SINGLEIS:
1268: case MAT_STRUCTURE_ONLY:
1269: /* These options are handled directly by MatSetOption() */
1270: break;
1271: default:
1272: SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_SUP,"unknown option %d",op);
1273: }
1274: return(0);
1275: }
1277: /* used for both SeqBAIJ and SeqSBAIJ matrices */
1278: PetscErrorCode MatGetRow_SeqBAIJ_private(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v,PetscInt *ai,PetscInt *aj,PetscScalar *aa)
1279: {
1281: PetscInt itmp,i,j,k,M,bn,bp,*idx_i,bs,bs2;
1282: MatScalar *aa_i;
1283: PetscScalar *v_i;
1286: bs = A->rmap->bs;
1287: bs2 = bs*bs;
1288: if (row < 0 || row >= A->rmap->N) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row %D out of range", row);
1290: bn = row/bs; /* Block number */
1291: bp = row % bs; /* Block Position */
1292: M = ai[bn+1] - ai[bn];
1293: *nz = bs*M;
1295: if (v) {
1296: *v = 0;
1297: if (*nz) {
1298: PetscMalloc1(*nz,v);
1299: for (i=0; i<M; i++) { /* for each block in the block row */
1300: v_i = *v + i*bs;
1301: aa_i = aa + bs2*(ai[bn] + i);
1302: for (j=bp,k=0; j<bs2; j+=bs,k++) v_i[k] = aa_i[j];
1303: }
1304: }
1305: }
1307: if (idx) {
1308: *idx = 0;
1309: if (*nz) {
1310: PetscMalloc1(*nz,idx);
1311: for (i=0; i<M; i++) { /* for each block in the block row */
1312: idx_i = *idx + i*bs;
1313: itmp = bs*aj[ai[bn] + i];
1314: for (j=0; j<bs; j++) idx_i[j] = itmp++;
1315: }
1316: }
1317: }
1318: return(0);
1319: }
1321: PetscErrorCode MatGetRow_SeqBAIJ(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v)
1322: {
1323: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
1327: MatGetRow_SeqBAIJ_private(A,row,nz,idx,v,a->i,a->j,a->a);
1328: return(0);
1329: }
1331: PetscErrorCode MatRestoreRow_SeqBAIJ(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v)
1332: {
1336: if (idx) {PetscFree(*idx);}
1337: if (v) {PetscFree(*v);}
1338: return(0);
1339: }
1341: PetscErrorCode MatTranspose_SeqBAIJ(Mat A,MatReuse reuse,Mat *B)
1342: {
1343: Mat_SeqBAIJ *a=(Mat_SeqBAIJ*)A->data;
1344: Mat C;
1346: PetscInt i,j,k,*aj=a->j,*ai=a->i,bs=A->rmap->bs,mbs=a->mbs,nbs=a->nbs,len,*col;
1347: PetscInt *rows,*cols,bs2=a->bs2;
1348: MatScalar *array;
1351: if (reuse == MAT_INITIAL_MATRIX || reuse == MAT_INPLACE_MATRIX) {
1352: PetscCalloc1(1+nbs,&col);
1354: for (i=0; i<ai[mbs]; i++) col[aj[i]] += 1;
1355: MatCreate(PetscObjectComm((PetscObject)A),&C);
1356: MatSetSizes(C,A->cmap->n,A->rmap->N,A->cmap->n,A->rmap->N);
1357: MatSetType(C,((PetscObject)A)->type_name);
1358: MatSeqBAIJSetPreallocation(C,bs,0,col);
1359: PetscFree(col);
1360: } else {
1361: C = *B;
1362: }
1364: array = a->a;
1365: PetscMalloc2(bs,&rows,bs,&cols);
1366: for (i=0; i<mbs; i++) {
1367: cols[0] = i*bs;
1368: for (k=1; k<bs; k++) cols[k] = cols[k-1] + 1;
1369: len = ai[i+1] - ai[i];
1370: for (j=0; j<len; j++) {
1371: rows[0] = (*aj++)*bs;
1372: for (k=1; k<bs; k++) rows[k] = rows[k-1] + 1;
1373: MatSetValues_SeqBAIJ(C,bs,rows,bs,cols,array,INSERT_VALUES);
1374: array += bs2;
1375: }
1376: }
1377: PetscFree2(rows,cols);
1379: MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);
1380: MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);
1382: if (reuse == MAT_INITIAL_MATRIX || reuse == MAT_REUSE_MATRIX) {
1383: *B = C;
1384: } else {
1385: MatHeaderMerge(A,&C);
1386: }
1387: return(0);
1388: }
1390: PetscErrorCode MatIsTranspose_SeqBAIJ(Mat A,Mat B,PetscReal tol,PetscBool *f)
1391: {
1393: Mat Btrans;
1396: *f = PETSC_FALSE;
1397: MatTranspose_SeqBAIJ(A,MAT_INITIAL_MATRIX,&Btrans);
1398: MatEqual_SeqBAIJ(B,Btrans,f);
1399: MatDestroy(&Btrans);
1400: return(0);
1401: }
1403: static PetscErrorCode MatView_SeqBAIJ_Binary(Mat A,PetscViewer viewer)
1404: {
1405: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
1407: PetscInt i,*col_lens,bs = A->rmap->bs,count,*jj,j,k,l,bs2=a->bs2;
1408: int fd;
1409: PetscScalar *aa;
1410: FILE *file;
1413: PetscViewerBinaryGetDescriptor(viewer,&fd);
1414: PetscMalloc1(4+A->rmap->N,&col_lens);
1415: col_lens[0] = MAT_FILE_CLASSID;
1417: col_lens[1] = A->rmap->N;
1418: col_lens[2] = A->cmap->n;
1419: col_lens[3] = a->nz*bs2;
1421: /* store lengths of each row and write (including header) to file */
1422: count = 0;
1423: for (i=0; i<a->mbs; i++) {
1424: for (j=0; j<bs; j++) {
1425: col_lens[4+count++] = bs*(a->i[i+1] - a->i[i]);
1426: }
1427: }
1428: PetscBinaryWrite(fd,col_lens,4+A->rmap->N,PETSC_INT,PETSC_TRUE);
1429: PetscFree(col_lens);
1431: /* store column indices (zero start index) */
1432: PetscMalloc1((a->nz+1)*bs2,&jj);
1433: count = 0;
1434: for (i=0; i<a->mbs; i++) {
1435: for (j=0; j<bs; j++) {
1436: for (k=a->i[i]; k<a->i[i+1]; k++) {
1437: for (l=0; l<bs; l++) {
1438: jj[count++] = bs*a->j[k] + l;
1439: }
1440: }
1441: }
1442: }
1443: PetscBinaryWrite(fd,jj,bs2*a->nz,PETSC_INT,PETSC_FALSE);
1444: PetscFree(jj);
1446: /* store nonzero values */
1447: PetscMalloc1((a->nz+1)*bs2,&aa);
1448: count = 0;
1449: for (i=0; i<a->mbs; i++) {
1450: for (j=0; j<bs; j++) {
1451: for (k=a->i[i]; k<a->i[i+1]; k++) {
1452: for (l=0; l<bs; l++) {
1453: aa[count++] = a->a[bs2*k + l*bs + j];
1454: }
1455: }
1456: }
1457: }
1458: PetscBinaryWrite(fd,aa,bs2*a->nz,PETSC_SCALAR,PETSC_FALSE);
1459: PetscFree(aa);
1461: PetscViewerBinaryGetInfoPointer(viewer,&file);
1462: if (file) {
1463: fprintf(file,"-matload_block_size %d\n",(int)A->rmap->bs);
1464: }
1465: return(0);
1466: }
1468: static PetscErrorCode MatView_SeqBAIJ_ASCII_structonly(Mat A,PetscViewer viewer)
1469: {
1471: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
1472: PetscInt i,bs = A->rmap->bs,k;
1475: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
1476: for (i=0; i<a->mbs; i++) {
1477: PetscViewerASCIIPrintf(viewer,"row %D-%D:",i*bs,i*bs+bs-1);
1478: for (k=a->i[i]; k<a->i[i+1]; k++) {
1479: PetscViewerASCIIPrintf(viewer," (%D-%D) ",bs*a->j[k],bs*a->j[k]+bs-1);
1480: }
1481: PetscViewerASCIIPrintf(viewer,"\n");
1482: }
1483: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
1484: return(0);
1485: }
1487: static PetscErrorCode MatView_SeqBAIJ_ASCII(Mat A,PetscViewer viewer)
1488: {
1489: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
1490: PetscErrorCode ierr;
1491: PetscInt i,j,bs = A->rmap->bs,k,l,bs2=a->bs2;
1492: PetscViewerFormat format;
1495: if (A->structure_only) {
1496: MatView_SeqBAIJ_ASCII_structonly(A,viewer);
1497: return(0);
1498: }
1500: PetscViewerGetFormat(viewer,&format);
1501: if (format == PETSC_VIEWER_ASCII_INFO || format == PETSC_VIEWER_ASCII_INFO_DETAIL) {
1502: PetscViewerASCIIPrintf(viewer," block size is %D\n",bs);
1503: } else if (format == PETSC_VIEWER_ASCII_MATLAB) {
1504: const char *matname;
1505: Mat aij;
1506: MatConvert(A,MATSEQAIJ,MAT_INITIAL_MATRIX,&aij);
1507: PetscObjectGetName((PetscObject)A,&matname);
1508: PetscObjectSetName((PetscObject)aij,matname);
1509: MatView(aij,viewer);
1510: MatDestroy(&aij);
1511: } else if (format == PETSC_VIEWER_ASCII_FACTOR_INFO) {
1512: return(0);
1513: } else if (format == PETSC_VIEWER_ASCII_COMMON) {
1514: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
1515: for (i=0; i<a->mbs; i++) {
1516: for (j=0; j<bs; j++) {
1517: PetscViewerASCIIPrintf(viewer,"row %D:",i*bs+j);
1518: for (k=a->i[i]; k<a->i[i+1]; k++) {
1519: for (l=0; l<bs; l++) {
1520: #if defined(PETSC_USE_COMPLEX)
1521: if (PetscImaginaryPart(a->a[bs2*k + l*bs + j]) > 0.0 && PetscRealPart(a->a[bs2*k + l*bs + j]) != 0.0) {
1522: PetscViewerASCIIPrintf(viewer," (%D, %g + %gi) ",bs*a->j[k]+l,
1523: (double)PetscRealPart(a->a[bs2*k + l*bs + j]),(double)PetscImaginaryPart(a->a[bs2*k + l*bs + j]));
1524: } else if (PetscImaginaryPart(a->a[bs2*k + l*bs + j]) < 0.0 && PetscRealPart(a->a[bs2*k + l*bs + j]) != 0.0) {
1525: PetscViewerASCIIPrintf(viewer," (%D, %g - %gi) ",bs*a->j[k]+l,
1526: (double)PetscRealPart(a->a[bs2*k + l*bs + j]),-(double)PetscImaginaryPart(a->a[bs2*k + l*bs + j]));
1527: } else if (PetscRealPart(a->a[bs2*k + l*bs + j]) != 0.0) {
1528: PetscViewerASCIIPrintf(viewer," (%D, %g) ",bs*a->j[k]+l,(double)PetscRealPart(a->a[bs2*k + l*bs + j]));
1529: }
1530: #else
1531: if (a->a[bs2*k + l*bs + j] != 0.0) {
1532: PetscViewerASCIIPrintf(viewer," (%D, %g) ",bs*a->j[k]+l,(double)a->a[bs2*k + l*bs + j]);
1533: }
1534: #endif
1535: }
1536: }
1537: PetscViewerASCIIPrintf(viewer,"\n");
1538: }
1539: }
1540: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
1541: } else {
1542: PetscViewerASCIIUseTabs(viewer,PETSC_FALSE);
1543: for (i=0; i<a->mbs; i++) {
1544: for (j=0; j<bs; j++) {
1545: PetscViewerASCIIPrintf(viewer,"row %D:",i*bs+j);
1546: for (k=a->i[i]; k<a->i[i+1]; k++) {
1547: for (l=0; l<bs; l++) {
1548: #if defined(PETSC_USE_COMPLEX)
1549: if (PetscImaginaryPart(a->a[bs2*k + l*bs + j]) > 0.0) {
1550: PetscViewerASCIIPrintf(viewer," (%D, %g + %g i) ",bs*a->j[k]+l,
1551: (double)PetscRealPart(a->a[bs2*k + l*bs + j]),(double)PetscImaginaryPart(a->a[bs2*k + l*bs + j]));
1552: } else if (PetscImaginaryPart(a->a[bs2*k + l*bs + j]) < 0.0) {
1553: PetscViewerASCIIPrintf(viewer," (%D, %g - %g i) ",bs*a->j[k]+l,
1554: (double)PetscRealPart(a->a[bs2*k + l*bs + j]),-(double)PetscImaginaryPart(a->a[bs2*k + l*bs + j]));
1555: } else {
1556: PetscViewerASCIIPrintf(viewer," (%D, %g) ",bs*a->j[k]+l,(double)PetscRealPart(a->a[bs2*k + l*bs + j]));
1557: }
1558: #else
1559: PetscViewerASCIIPrintf(viewer," (%D, %g) ",bs*a->j[k]+l,(double)a->a[bs2*k + l*bs + j]);
1560: #endif
1561: }
1562: }
1563: PetscViewerASCIIPrintf(viewer,"\n");
1564: }
1565: }
1566: PetscViewerASCIIUseTabs(viewer,PETSC_TRUE);
1567: }
1568: PetscViewerFlush(viewer);
1569: return(0);
1570: }
1572: #include <petscdraw.h>
1573: static PetscErrorCode MatView_SeqBAIJ_Draw_Zoom(PetscDraw draw,void *Aa)
1574: {
1575: Mat A = (Mat) Aa;
1576: Mat_SeqBAIJ *a=(Mat_SeqBAIJ*)A->data;
1577: PetscErrorCode ierr;
1578: PetscInt row,i,j,k,l,mbs=a->mbs,color,bs=A->rmap->bs,bs2=a->bs2;
1579: PetscReal xl,yl,xr,yr,x_l,x_r,y_l,y_r;
1580: MatScalar *aa;
1581: PetscViewer viewer;
1582: PetscViewerFormat format;
1585: PetscObjectQuery((PetscObject)A,"Zoomviewer",(PetscObject*)&viewer);
1586: PetscViewerGetFormat(viewer,&format);
1587: PetscDrawGetCoordinates(draw,&xl,&yl,&xr,&yr);
1589: /* loop over matrix elements drawing boxes */
1591: if (format != PETSC_VIEWER_DRAW_CONTOUR) {
1592: PetscDrawCollectiveBegin(draw);
1593: /* Blue for negative, Cyan for zero and Red for positive */
1594: color = PETSC_DRAW_BLUE;
1595: for (i=0,row=0; i<mbs; i++,row+=bs) {
1596: for (j=a->i[i]; j<a->i[i+1]; j++) {
1597: y_l = A->rmap->N - row - 1.0; y_r = y_l + 1.0;
1598: x_l = a->j[j]*bs; x_r = x_l + 1.0;
1599: aa = a->a + j*bs2;
1600: for (k=0; k<bs; k++) {
1601: for (l=0; l<bs; l++) {
1602: if (PetscRealPart(*aa++) >= 0.) continue;
1603: PetscDrawRectangle(draw,x_l+k,y_l-l,x_r+k,y_r-l,color,color,color,color);
1604: }
1605: }
1606: }
1607: }
1608: color = PETSC_DRAW_CYAN;
1609: for (i=0,row=0; i<mbs; i++,row+=bs) {
1610: for (j=a->i[i]; j<a->i[i+1]; j++) {
1611: y_l = A->rmap->N - row - 1.0; y_r = y_l + 1.0;
1612: x_l = a->j[j]*bs; x_r = x_l + 1.0;
1613: aa = a->a + j*bs2;
1614: for (k=0; k<bs; k++) {
1615: for (l=0; l<bs; l++) {
1616: if (PetscRealPart(*aa++) != 0.) continue;
1617: PetscDrawRectangle(draw,x_l+k,y_l-l,x_r+k,y_r-l,color,color,color,color);
1618: }
1619: }
1620: }
1621: }
1622: color = PETSC_DRAW_RED;
1623: for (i=0,row=0; i<mbs; i++,row+=bs) {
1624: for (j=a->i[i]; j<a->i[i+1]; j++) {
1625: y_l = A->rmap->N - row - 1.0; y_r = y_l + 1.0;
1626: x_l = a->j[j]*bs; x_r = x_l + 1.0;
1627: aa = a->a + j*bs2;
1628: for (k=0; k<bs; k++) {
1629: for (l=0; l<bs; l++) {
1630: if (PetscRealPart(*aa++) <= 0.) continue;
1631: PetscDrawRectangle(draw,x_l+k,y_l-l,x_r+k,y_r-l,color,color,color,color);
1632: }
1633: }
1634: }
1635: }
1636: PetscDrawCollectiveEnd(draw);
1637: } else {
1638: /* use contour shading to indicate magnitude of values */
1639: /* first determine max of all nonzero values */
1640: PetscReal minv = 0.0, maxv = 0.0;
1641: PetscDraw popup;
1643: for (i=0; i<a->nz*a->bs2; i++) {
1644: if (PetscAbsScalar(a->a[i]) > maxv) maxv = PetscAbsScalar(a->a[i]);
1645: }
1646: if (minv >= maxv) maxv = minv + PETSC_SMALL;
1647: PetscDrawGetPopup(draw,&popup);
1648: PetscDrawScalePopup(popup,0.0,maxv);
1650: PetscDrawCollectiveBegin(draw);
1651: for (i=0,row=0; i<mbs; i++,row+=bs) {
1652: for (j=a->i[i]; j<a->i[i+1]; j++) {
1653: y_l = A->rmap->N - row - 1.0; y_r = y_l + 1.0;
1654: x_l = a->j[j]*bs; x_r = x_l + 1.0;
1655: aa = a->a + j*bs2;
1656: for (k=0; k<bs; k++) {
1657: for (l=0; l<bs; l++) {
1658: MatScalar v = *aa++;
1659: color = PetscDrawRealToColor(PetscAbsScalar(v),minv,maxv);
1660: PetscDrawRectangle(draw,x_l+k,y_l-l,x_r+k,y_r-l,color,color,color,color);
1661: }
1662: }
1663: }
1664: }
1665: PetscDrawCollectiveEnd(draw);
1666: }
1667: return(0);
1668: }
1670: static PetscErrorCode MatView_SeqBAIJ_Draw(Mat A,PetscViewer viewer)
1671: {
1673: PetscReal xl,yl,xr,yr,w,h;
1674: PetscDraw draw;
1675: PetscBool isnull;
1678: PetscViewerDrawGetDraw(viewer,0,&draw);
1679: PetscDrawIsNull(draw,&isnull);
1680: if (isnull) return(0);
1682: xr = A->cmap->n; yr = A->rmap->N; h = yr/10.0; w = xr/10.0;
1683: xr += w; yr += h; xl = -w; yl = -h;
1684: PetscDrawSetCoordinates(draw,xl,yl,xr,yr);
1685: PetscObjectCompose((PetscObject)A,"Zoomviewer",(PetscObject)viewer);
1686: PetscDrawZoom(draw,MatView_SeqBAIJ_Draw_Zoom,A);
1687: PetscObjectCompose((PetscObject)A,"Zoomviewer",NULL);
1688: PetscDrawSave(draw);
1689: return(0);
1690: }
1692: PetscErrorCode MatView_SeqBAIJ(Mat A,PetscViewer viewer)
1693: {
1695: PetscBool iascii,isbinary,isdraw;
1698: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);
1699: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary);
1700: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERDRAW,&isdraw);
1701: if (iascii) {
1702: MatView_SeqBAIJ_ASCII(A,viewer);
1703: } else if (isbinary) {
1704: MatView_SeqBAIJ_Binary(A,viewer);
1705: } else if (isdraw) {
1706: MatView_SeqBAIJ_Draw(A,viewer);
1707: } else {
1708: Mat B;
1709: MatConvert(A,MATSEQAIJ,MAT_INITIAL_MATRIX,&B);
1710: MatView(B,viewer);
1711: MatDestroy(&B);
1712: }
1713: return(0);
1714: }
1717: PetscErrorCode MatGetValues_SeqBAIJ(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],PetscScalar v[])
1718: {
1719: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
1720: PetscInt *rp,k,low,high,t,row,nrow,i,col,l,*aj = a->j;
1721: PetscInt *ai = a->i,*ailen = a->ilen;
1722: PetscInt brow,bcol,ridx,cidx,bs=A->rmap->bs,bs2=a->bs2;
1723: MatScalar *ap,*aa = a->a;
1726: for (k=0; k<m; k++) { /* loop over rows */
1727: row = im[k]; brow = row/bs;
1728: if (row < 0) {v += n; continue;} /* SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative row"); */
1729: if (row >= A->rmap->N) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row %D too large", row);
1730: rp = aj + ai[brow]; ap = aa + bs2*ai[brow];
1731: nrow = ailen[brow];
1732: for (l=0; l<n; l++) { /* loop over columns */
1733: if (in[l] < 0) {v++; continue;} /* SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative column"); */
1734: if (in[l] >= A->cmap->n) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column %D too large", in[l]);
1735: col = in[l];
1736: bcol = col/bs;
1737: cidx = col%bs;
1738: ridx = row%bs;
1739: high = nrow;
1740: low = 0; /* assume unsorted */
1741: while (high-low > 5) {
1742: t = (low+high)/2;
1743: if (rp[t] > bcol) high = t;
1744: else low = t;
1745: }
1746: for (i=low; i<high; i++) {
1747: if (rp[i] > bcol) break;
1748: if (rp[i] == bcol) {
1749: *v++ = ap[bs2*i+bs*cidx+ridx];
1750: goto finished;
1751: }
1752: }
1753: *v++ = 0.0;
1754: finished:;
1755: }
1756: }
1757: return(0);
1758: }
1760: PetscErrorCode MatSetValuesBlocked_SeqBAIJ(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode is)
1761: {
1762: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
1763: PetscInt *rp,k,low,high,t,ii,jj,row,nrow,i,col,l,rmax,N,lastcol = -1;
1764: PetscInt *imax=a->imax,*ai=a->i,*ailen=a->ilen;
1765: PetscErrorCode ierr;
1766: PetscInt *aj =a->j,nonew=a->nonew,bs2=a->bs2,bs=A->rmap->bs,stepval;
1767: PetscBool roworiented=a->roworiented;
1768: const PetscScalar *value = v;
1769: MatScalar *ap=NULL,*aa = a->a,*bap;
1772: if (roworiented) {
1773: stepval = (n-1)*bs;
1774: } else {
1775: stepval = (m-1)*bs;
1776: }
1777: for (k=0; k<m; k++) { /* loop over added rows */
1778: row = im[k];
1779: if (row < 0) continue;
1780: #if defined(PETSC_USE_DEBUG)
1781: if (row >= a->mbs) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Block row index too large %D max %D",row,a->mbs-1);
1782: #endif
1783: rp = aj + ai[row];
1784: if (!A->structure_only) ap = aa + bs2*ai[row];
1785: rmax = imax[row];
1786: nrow = ailen[row];
1787: low = 0;
1788: high = nrow;
1789: for (l=0; l<n; l++) { /* loop over added columns */
1790: if (in[l] < 0) continue;
1791: #if defined(PETSC_USE_DEBUG)
1792: if (in[l] >= a->nbs) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Block column index too large %D max %D",in[l],a->nbs-1);
1793: #endif
1794: col = in[l];
1795: if (!A->structure_only) {
1796: if (roworiented) {
1797: value = v + (k*(stepval+bs) + l)*bs;
1798: } else {
1799: value = v + (l*(stepval+bs) + k)*bs;
1800: }
1801: }
1802: if (col <= lastcol) low = 0;
1803: else high = nrow;
1804: lastcol = col;
1805: while (high-low > 7) {
1806: t = (low+high)/2;
1807: if (rp[t] > col) high = t;
1808: else low = t;
1809: }
1810: for (i=low; i<high; i++) {
1811: if (rp[i] > col) break;
1812: if (rp[i] == col) {
1813: if (A->structure_only) goto noinsert2;
1814: bap = ap + bs2*i;
1815: if (roworiented) {
1816: if (is == ADD_VALUES) {
1817: for (ii=0; ii<bs; ii++,value+=stepval) {
1818: for (jj=ii; jj<bs2; jj+=bs) {
1819: bap[jj] += *value++;
1820: }
1821: }
1822: } else {
1823: for (ii=0; ii<bs; ii++,value+=stepval) {
1824: for (jj=ii; jj<bs2; jj+=bs) {
1825: bap[jj] = *value++;
1826: }
1827: }
1828: }
1829: } else {
1830: if (is == ADD_VALUES) {
1831: for (ii=0; ii<bs; ii++,value+=bs+stepval) {
1832: for (jj=0; jj<bs; jj++) {
1833: bap[jj] += value[jj];
1834: }
1835: bap += bs;
1836: }
1837: } else {
1838: for (ii=0; ii<bs; ii++,value+=bs+stepval) {
1839: for (jj=0; jj<bs; jj++) {
1840: bap[jj] = value[jj];
1841: }
1842: bap += bs;
1843: }
1844: }
1845: }
1846: goto noinsert2;
1847: }
1848: }
1849: if (nonew == 1) goto noinsert2;
1850: if (nonew == -1) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new blocked index new nonzero block (%D, %D) in the matrix", row, col);
1851: if (A->structure_only) {
1852: MatSeqXAIJReallocateAIJ_structure_only(A,a->mbs,bs2,nrow,row,col,rmax,ai,aj,rp,imax,nonew,MatScalar);
1853: } else {
1854: MatSeqXAIJReallocateAIJ(A,a->mbs,bs2,nrow,row,col,rmax,aa,ai,aj,rp,ap,imax,nonew,MatScalar);
1855: }
1856: N = nrow++ - 1; high++;
1857: /* shift up all the later entries in this row */
1858: for (ii=N; ii>=i; ii--) {
1859: rp[ii+1] = rp[ii];
1860: if (!A->structure_only) {
1861: PetscMemcpy(ap+bs2*(ii+1),ap+bs2*(ii),bs2*sizeof(MatScalar));
1862: }
1863: }
1864: if (N >= i && !A->structure_only) {
1865: PetscMemzero(ap+bs2*i,bs2*sizeof(MatScalar));
1866: }
1868: rp[i] = col;
1869: if (!A->structure_only) {
1870: bap = ap + bs2*i;
1871: if (roworiented) {
1872: for (ii=0; ii<bs; ii++,value+=stepval) {
1873: for (jj=ii; jj<bs2; jj+=bs) {
1874: bap[jj] = *value++;
1875: }
1876: }
1877: } else {
1878: for (ii=0; ii<bs; ii++,value+=stepval) {
1879: for (jj=0; jj<bs; jj++) {
1880: *bap++ = *value++;
1881: }
1882: }
1883: }
1884: }
1885: noinsert2:;
1886: low = i;
1887: }
1888: ailen[row] = nrow;
1889: }
1890: return(0);
1891: }
1893: PetscErrorCode MatAssemblyEnd_SeqBAIJ(Mat A,MatAssemblyType mode)
1894: {
1895: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
1896: PetscInt fshift = 0,i,j,*ai = a->i,*aj = a->j,*imax = a->imax;
1897: PetscInt m = A->rmap->N,*ip,N,*ailen = a->ilen;
1899: PetscInt mbs = a->mbs,bs2 = a->bs2,rmax = 0;
1900: MatScalar *aa = a->a,*ap;
1901: PetscReal ratio=0.6;
1904: if (mode == MAT_FLUSH_ASSEMBLY) return(0);
1906: if (m) rmax = ailen[0];
1907: for (i=1; i<mbs; i++) {
1908: /* move each row back by the amount of empty slots (fshift) before it*/
1909: fshift += imax[i-1] - ailen[i-1];
1910: rmax = PetscMax(rmax,ailen[i]);
1911: if (fshift) {
1912: ip = aj + ai[i]; ap = aa + bs2*ai[i];
1913: N = ailen[i];
1914: for (j=0; j<N; j++) {
1915: ip[j-fshift] = ip[j];
1916: if (!A->structure_only) {
1917: PetscMemcpy(ap+(j-fshift)*bs2,ap+j*bs2,bs2*sizeof(MatScalar));
1918: }
1919: }
1920: }
1921: ai[i] = ai[i-1] + ailen[i-1];
1922: }
1923: if (mbs) {
1924: fshift += imax[mbs-1] - ailen[mbs-1];
1925: ai[mbs] = ai[mbs-1] + ailen[mbs-1];
1926: }
1928: /* reset ilen and imax for each row */
1929: a->nonzerorowcnt = 0;
1930: if (A->structure_only) {
1931: PetscFree2(a->imax,a->ilen);
1932: } else { /* !A->structure_only */
1933: for (i=0; i<mbs; i++) {
1934: ailen[i] = imax[i] = ai[i+1] - ai[i];
1935: a->nonzerorowcnt += ((ai[i+1] - ai[i]) > 0);
1936: }
1937: }
1938: a->nz = ai[mbs];
1940: /* diagonals may have moved, so kill the diagonal pointers */
1941: a->idiagvalid = PETSC_FALSE;
1942: if (fshift && a->diag) {
1943: PetscFree(a->diag);
1944: PetscLogObjectMemory((PetscObject)A,-(mbs+1)*sizeof(PetscInt));
1945: a->diag = 0;
1946: }
1947: if (fshift && a->nounused == -1) SETERRQ4(PETSC_COMM_SELF,PETSC_ERR_PLIB, "Unused space detected in matrix: %D X %D block size %D, %D unneeded", m, A->cmap->n, A->rmap->bs, fshift*bs2);
1948: PetscInfo5(A,"Matrix size: %D X %D, block size %D; storage space: %D unneeded, %D used\n",m,A->cmap->n,A->rmap->bs,fshift*bs2,a->nz*bs2);
1949: PetscInfo1(A,"Number of mallocs during MatSetValues is %D\n",a->reallocs);
1950: PetscInfo1(A,"Most nonzeros blocks in any row is %D\n",rmax);
1952: A->info.mallocs += a->reallocs;
1953: a->reallocs = 0;
1954: A->info.nz_unneeded = (PetscReal)fshift*bs2;
1955: a->rmax = rmax;
1957: if (!A->structure_only) {
1958: MatCheckCompressedRow(A,a->nonzerorowcnt,&a->compressedrow,a->i,mbs,ratio);
1959: }
1960: return(0);
1961: }
1963: /*
1964: This function returns an array of flags which indicate the locations of contiguous
1965: blocks that should be zeroed. for eg: if bs = 3 and is = [0,1,2,3,5,6,7,8,9]
1966: then the resulting sizes = [3,1,1,3,1] correspondig to sets [(0,1,2),(3),(5),(6,7,8),(9)]
1967: Assume: sizes should be long enough to hold all the values.
1968: */
1969: static PetscErrorCode MatZeroRows_SeqBAIJ_Check_Blocks(PetscInt idx[],PetscInt n,PetscInt bs,PetscInt sizes[], PetscInt *bs_max)
1970: {
1971: PetscInt i,j,k,row;
1972: PetscBool flg;
1975: for (i=0,j=0; i<n; j++) {
1976: row = idx[i];
1977: if (row%bs!=0) { /* Not the begining of a block */
1978: sizes[j] = 1;
1979: i++;
1980: } else if (i+bs > n) { /* complete block doesn't exist (at idx end) */
1981: sizes[j] = 1; /* Also makes sure atleast 'bs' values exist for next else */
1982: i++;
1983: } else { /* Begining of the block, so check if the complete block exists */
1984: flg = PETSC_TRUE;
1985: for (k=1; k<bs; k++) {
1986: if (row+k != idx[i+k]) { /* break in the block */
1987: flg = PETSC_FALSE;
1988: break;
1989: }
1990: }
1991: if (flg) { /* No break in the bs */
1992: sizes[j] = bs;
1993: i += bs;
1994: } else {
1995: sizes[j] = 1;
1996: i++;
1997: }
1998: }
1999: }
2000: *bs_max = j;
2001: return(0);
2002: }
2004: PetscErrorCode MatZeroRows_SeqBAIJ(Mat A,PetscInt is_n,const PetscInt is_idx[],PetscScalar diag,Vec x, Vec b)
2005: {
2006: Mat_SeqBAIJ *baij=(Mat_SeqBAIJ*)A->data;
2007: PetscErrorCode ierr;
2008: PetscInt i,j,k,count,*rows;
2009: PetscInt bs=A->rmap->bs,bs2=baij->bs2,*sizes,row,bs_max;
2010: PetscScalar zero = 0.0;
2011: MatScalar *aa;
2012: const PetscScalar *xx;
2013: PetscScalar *bb;
2016: /* fix right hand side if needed */
2017: if (x && b) {
2018: VecGetArrayRead(x,&xx);
2019: VecGetArray(b,&bb);
2020: for (i=0; i<is_n; i++) {
2021: bb[is_idx[i]] = diag*xx[is_idx[i]];
2022: }
2023: VecRestoreArrayRead(x,&xx);
2024: VecRestoreArray(b,&bb);
2025: }
2027: /* Make a copy of the IS and sort it */
2028: /* allocate memory for rows,sizes */
2029: PetscMalloc2(is_n,&rows,2*is_n,&sizes);
2031: /* copy IS values to rows, and sort them */
2032: for (i=0; i<is_n; i++) rows[i] = is_idx[i];
2033: PetscSortInt(is_n,rows);
2035: if (baij->keepnonzeropattern) {
2036: for (i=0; i<is_n; i++) sizes[i] = 1;
2037: bs_max = is_n;
2038: } else {
2039: MatZeroRows_SeqBAIJ_Check_Blocks(rows,is_n,bs,sizes,&bs_max);
2040: A->nonzerostate++;
2041: }
2043: for (i=0,j=0; i<bs_max; j+=sizes[i],i++) {
2044: row = rows[j];
2045: if (row < 0 || row > A->rmap->N) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range",row);
2046: count = (baij->i[row/bs +1] - baij->i[row/bs])*bs;
2047: aa = ((MatScalar*)(baij->a)) + baij->i[row/bs]*bs2 + (row%bs);
2048: if (sizes[i] == bs && !baij->keepnonzeropattern) {
2049: if (diag != (PetscScalar)0.0) {
2050: if (baij->ilen[row/bs] > 0) {
2051: baij->ilen[row/bs] = 1;
2052: baij->j[baij->i[row/bs]] = row/bs;
2054: PetscMemzero(aa,count*bs*sizeof(MatScalar));
2055: }
2056: /* Now insert all the diagonal values for this bs */
2057: for (k=0; k<bs; k++) {
2058: (*A->ops->setvalues)(A,1,rows+j+k,1,rows+j+k,&diag,INSERT_VALUES);
2059: }
2060: } else { /* (diag == 0.0) */
2061: baij->ilen[row/bs] = 0;
2062: } /* end (diag == 0.0) */
2063: } else { /* (sizes[i] != bs) */
2064: #if defined(PETSC_USE_DEBUG)
2065: if (sizes[i] != 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Internal Error. Value should be 1");
2066: #endif
2067: for (k=0; k<count; k++) {
2068: aa[0] = zero;
2069: aa += bs;
2070: }
2071: if (diag != (PetscScalar)0.0) {
2072: (*A->ops->setvalues)(A,1,rows+j,1,rows+j,&diag,INSERT_VALUES);
2073: }
2074: }
2075: }
2077: PetscFree2(rows,sizes);
2078: MatAssemblyEnd_SeqBAIJ(A,MAT_FINAL_ASSEMBLY);
2079: return(0);
2080: }
2082: PetscErrorCode MatZeroRowsColumns_SeqBAIJ(Mat A,PetscInt is_n,const PetscInt is_idx[],PetscScalar diag,Vec x, Vec b)
2083: {
2084: Mat_SeqBAIJ *baij=(Mat_SeqBAIJ*)A->data;
2085: PetscErrorCode ierr;
2086: PetscInt i,j,k,count;
2087: PetscInt bs =A->rmap->bs,bs2=baij->bs2,row,col;
2088: PetscScalar zero = 0.0;
2089: MatScalar *aa;
2090: const PetscScalar *xx;
2091: PetscScalar *bb;
2092: PetscBool *zeroed,vecs = PETSC_FALSE;
2095: /* fix right hand side if needed */
2096: if (x && b) {
2097: VecGetArrayRead(x,&xx);
2098: VecGetArray(b,&bb);
2099: vecs = PETSC_TRUE;
2100: }
2102: /* zero the columns */
2103: PetscCalloc1(A->rmap->n,&zeroed);
2104: for (i=0; i<is_n; i++) {
2105: if (is_idx[i] < 0 || is_idx[i] >= A->rmap->N) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range",is_idx[i]);
2106: zeroed[is_idx[i]] = PETSC_TRUE;
2107: }
2108: for (i=0; i<A->rmap->N; i++) {
2109: if (!zeroed[i]) {
2110: row = i/bs;
2111: for (j=baij->i[row]; j<baij->i[row+1]; j++) {
2112: for (k=0; k<bs; k++) {
2113: col = bs*baij->j[j] + k;
2114: if (zeroed[col]) {
2115: aa = ((MatScalar*)(baij->a)) + j*bs2 + (i%bs) + bs*k;
2116: if (vecs) bb[i] -= aa[0]*xx[col];
2117: aa[0] = 0.0;
2118: }
2119: }
2120: }
2121: } else if (vecs) bb[i] = diag*xx[i];
2122: }
2123: PetscFree(zeroed);
2124: if (vecs) {
2125: VecRestoreArrayRead(x,&xx);
2126: VecRestoreArray(b,&bb);
2127: }
2129: /* zero the rows */
2130: for (i=0; i<is_n; i++) {
2131: row = is_idx[i];
2132: count = (baij->i[row/bs +1] - baij->i[row/bs])*bs;
2133: aa = ((MatScalar*)(baij->a)) + baij->i[row/bs]*bs2 + (row%bs);
2134: for (k=0; k<count; k++) {
2135: aa[0] = zero;
2136: aa += bs;
2137: }
2138: if (diag != (PetscScalar)0.0) {
2139: (*A->ops->setvalues)(A,1,&row,1,&row,&diag,INSERT_VALUES);
2140: }
2141: }
2142: MatAssemblyEnd_SeqBAIJ(A,MAT_FINAL_ASSEMBLY);
2143: return(0);
2144: }
2146: PetscErrorCode MatSetValues_SeqBAIJ(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode is)
2147: {
2148: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
2149: PetscInt *rp,k,low,high,t,ii,row,nrow,i,col,l,rmax,N,lastcol = -1;
2150: PetscInt *imax=a->imax,*ai=a->i,*ailen=a->ilen;
2151: PetscInt *aj =a->j,nonew=a->nonew,bs=A->rmap->bs,brow,bcol;
2153: PetscInt ridx,cidx,bs2=a->bs2;
2154: PetscBool roworiented=a->roworiented;
2155: MatScalar *ap=NULL,value=0.0,*aa=a->a,*bap;
2158: for (k=0; k<m; k++) { /* loop over added rows */
2159: row = im[k];
2160: brow = row/bs;
2161: if (row < 0) continue;
2162: #if defined(PETSC_USE_DEBUG)
2163: if (row >= A->rmap->N) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",row,A->rmap->N-1);
2164: #endif
2165: rp = aj + ai[brow];
2166: if (!A->structure_only) ap = aa + bs2*ai[brow];
2167: rmax = imax[brow];
2168: nrow = ailen[brow];
2169: low = 0;
2170: high = nrow;
2171: for (l=0; l<n; l++) { /* loop over added columns */
2172: if (in[l] < 0) continue;
2173: #if defined(PETSC_USE_DEBUG)
2174: if (in[l] >= A->cmap->n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",in[l],A->cmap->n-1);
2175: #endif
2176: col = in[l]; bcol = col/bs;
2177: ridx = row % bs; cidx = col % bs;
2178: if (!A->structure_only) {
2179: if (roworiented) {
2180: value = v[l + k*n];
2181: } else {
2182: value = v[k + l*m];
2183: }
2184: }
2185: if (col <= lastcol) low = 0; else high = nrow;
2186: lastcol = col;
2187: while (high-low > 7) {
2188: t = (low+high)/2;
2189: if (rp[t] > bcol) high = t;
2190: else low = t;
2191: }
2192: for (i=low; i<high; i++) {
2193: if (rp[i] > bcol) break;
2194: if (rp[i] == bcol) {
2195: bap = ap + bs2*i + bs*cidx + ridx;
2196: if (!A->structure_only) {
2197: if (is == ADD_VALUES) *bap += value;
2198: else *bap = value;
2199: }
2200: goto noinsert1;
2201: }
2202: }
2203: if (nonew == 1) goto noinsert1;
2204: if (nonew == -1) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero (%D, %D) in the matrix", row, col);
2205: if (A->structure_only) {
2206: MatSeqXAIJReallocateAIJ_structure_only(A,a->mbs,bs2,nrow,brow,bcol,rmax,ai,aj,rp,imax,nonew,MatScalar);
2207: } else {
2208: MatSeqXAIJReallocateAIJ(A,a->mbs,bs2,nrow,brow,bcol,rmax,aa,ai,aj,rp,ap,imax,nonew,MatScalar);
2209: }
2210: N = nrow++ - 1; high++;
2211: /* shift up all the later entries in this row */
2212: for (ii=N; ii>=i; ii--) {
2213: rp[ii+1] = rp[ii];
2214: if (!A->structure_only) {
2215: PetscMemcpy(ap+bs2*(ii+1),ap+bs2*(ii),bs2*sizeof(MatScalar));
2216: }
2217: }
2218: if (N>=i && !A->structure_only) {
2219: PetscMemzero(ap+bs2*i,bs2*sizeof(MatScalar));
2220: }
2221: rp[i] = bcol;
2222: if (!A->structure_only) ap[bs2*i + bs*cidx + ridx] = value;
2223: a->nz++;
2224: A->nonzerostate++;
2225: noinsert1:;
2226: low = i;
2227: }
2228: ailen[brow] = nrow;
2229: }
2230: return(0);
2231: }
2233: PetscErrorCode MatILUFactor_SeqBAIJ(Mat inA,IS row,IS col,const MatFactorInfo *info)
2234: {
2235: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)inA->data;
2236: Mat outA;
2238: PetscBool row_identity,col_identity;
2241: if (info->levels != 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Only levels = 0 supported for in-place ILU");
2242: ISIdentity(row,&row_identity);
2243: ISIdentity(col,&col_identity);
2244: if (!row_identity || !col_identity) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Row and column permutations must be identity for in-place ILU");
2246: outA = inA;
2247: inA->factortype = MAT_FACTOR_LU;
2248: PetscFree(inA->solvertype);
2249: PetscStrallocpy(MATSOLVERPETSC,&inA->solvertype);
2251: MatMarkDiagonal_SeqBAIJ(inA);
2253: PetscObjectReference((PetscObject)row);
2254: ISDestroy(&a->row);
2255: a->row = row;
2256: PetscObjectReference((PetscObject)col);
2257: ISDestroy(&a->col);
2258: a->col = col;
2260: /* Create the invert permutation so that it can be used in MatLUFactorNumeric() */
2261: ISDestroy(&a->icol);
2262: ISInvertPermutation(col,PETSC_DECIDE,&a->icol);
2263: PetscLogObjectParent((PetscObject)inA,(PetscObject)a->icol);
2265: MatSeqBAIJSetNumericFactorization_inplace(inA,(PetscBool)(row_identity && col_identity));
2266: if (!a->solve_work) {
2267: PetscMalloc1(inA->rmap->N+inA->rmap->bs,&a->solve_work);
2268: PetscLogObjectMemory((PetscObject)inA,(inA->rmap->N+inA->rmap->bs)*sizeof(PetscScalar));
2269: }
2270: MatLUFactorNumeric(outA,inA,info);
2271: return(0);
2272: }
2274: PetscErrorCode MatSeqBAIJSetColumnIndices_SeqBAIJ(Mat mat,PetscInt *indices)
2275: {
2276: Mat_SeqBAIJ *baij = (Mat_SeqBAIJ*)mat->data;
2277: PetscInt i,nz,mbs;
2280: nz = baij->maxnz;
2281: mbs = baij->mbs;
2282: for (i=0; i<nz; i++) {
2283: baij->j[i] = indices[i];
2284: }
2285: baij->nz = nz;
2286: for (i=0; i<mbs; i++) {
2287: baij->ilen[i] = baij->imax[i];
2288: }
2289: return(0);
2290: }
2292: /*@
2293: MatSeqBAIJSetColumnIndices - Set the column indices for all the rows
2294: in the matrix.
2296: Input Parameters:
2297: + mat - the SeqBAIJ matrix
2298: - indices - the column indices
2300: Level: advanced
2302: Notes:
2303: This can be called if you have precomputed the nonzero structure of the
2304: matrix and want to provide it to the matrix object to improve the performance
2305: of the MatSetValues() operation.
2307: You MUST have set the correct numbers of nonzeros per row in the call to
2308: MatCreateSeqBAIJ(), and the columns indices MUST be sorted.
2310: MUST be called before any calls to MatSetValues();
2312: @*/
2313: PetscErrorCode MatSeqBAIJSetColumnIndices(Mat mat,PetscInt *indices)
2314: {
2320: PetscUseMethod(mat,"MatSeqBAIJSetColumnIndices_C",(Mat,PetscInt*),(mat,indices));
2321: return(0);
2322: }
2324: PetscErrorCode MatGetRowMaxAbs_SeqBAIJ(Mat A,Vec v,PetscInt idx[])
2325: {
2326: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
2328: PetscInt i,j,n,row,bs,*ai,*aj,mbs;
2329: PetscReal atmp;
2330: PetscScalar *x,zero = 0.0;
2331: MatScalar *aa;
2332: PetscInt ncols,brow,krow,kcol;
2335: if (A->factortype) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
2336: bs = A->rmap->bs;
2337: aa = a->a;
2338: ai = a->i;
2339: aj = a->j;
2340: mbs = a->mbs;
2342: VecSet(v,zero);
2343: VecGetArray(v,&x);
2344: VecGetLocalSize(v,&n);
2345: if (n != A->rmap->N) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector");
2346: for (i=0; i<mbs; i++) {
2347: ncols = ai[1] - ai[0]; ai++;
2348: brow = bs*i;
2349: for (j=0; j<ncols; j++) {
2350: for (kcol=0; kcol<bs; kcol++) {
2351: for (krow=0; krow<bs; krow++) {
2352: atmp = PetscAbsScalar(*aa);aa++;
2353: row = brow + krow; /* row index */
2354: if (PetscAbsScalar(x[row]) < atmp) {x[row] = atmp; if (idx) idx[row] = bs*(*aj) + kcol;}
2355: }
2356: }
2357: aj++;
2358: }
2359: }
2360: VecRestoreArray(v,&x);
2361: return(0);
2362: }
2364: PetscErrorCode MatCopy_SeqBAIJ(Mat A,Mat B,MatStructure str)
2365: {
2369: /* If the two matrices have the same copy implementation, use fast copy. */
2370: if (str == SAME_NONZERO_PATTERN && (A->ops->copy == B->ops->copy)) {
2371: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
2372: Mat_SeqBAIJ *b = (Mat_SeqBAIJ*)B->data;
2373: PetscInt ambs=a->mbs,bmbs=b->mbs,abs=A->rmap->bs,bbs=B->rmap->bs,bs2=abs*abs;
2375: if (a->i[ambs] != b->i[bmbs]) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"Number of nonzero blocks in matrices A %D and B %D are different",a->i[ambs],b->i[bmbs]);
2376: if (abs != bbs) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"Block size A %D and B %D are different",abs,bbs);
2377: PetscMemcpy(b->a,a->a,(bs2*a->i[ambs])*sizeof(PetscScalar));
2378: PetscObjectStateIncrease((PetscObject)B);
2379: } else {
2380: MatCopy_Basic(A,B,str);
2381: }
2382: return(0);
2383: }
2385: PetscErrorCode MatSetUp_SeqBAIJ(Mat A)
2386: {
2390: MatSeqBAIJSetPreallocation(A,A->rmap->bs,PETSC_DEFAULT,0);
2391: return(0);
2392: }
2394: PetscErrorCode MatSeqBAIJGetArray_SeqBAIJ(Mat A,PetscScalar *array[])
2395: {
2396: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
2399: *array = a->a;
2400: return(0);
2401: }
2403: PetscErrorCode MatSeqBAIJRestoreArray_SeqBAIJ(Mat A,PetscScalar *array[])
2404: {
2406: return(0);
2407: }
2409: PetscErrorCode MatAXPYGetPreallocation_SeqBAIJ(Mat Y,Mat X,PetscInt *nnz)
2410: {
2411: PetscInt bs = Y->rmap->bs,mbs = Y->rmap->N/bs;
2412: Mat_SeqBAIJ *x = (Mat_SeqBAIJ*)X->data;
2413: Mat_SeqBAIJ *y = (Mat_SeqBAIJ*)Y->data;
2417: /* Set the number of nonzeros in the new matrix */
2418: MatAXPYGetPreallocation_SeqX_private(mbs,x->i,x->j,y->i,y->j,nnz);
2419: return(0);
2420: }
2422: PetscErrorCode MatAXPY_SeqBAIJ(Mat Y,PetscScalar a,Mat X,MatStructure str)
2423: {
2424: Mat_SeqBAIJ *x = (Mat_SeqBAIJ*)X->data,*y = (Mat_SeqBAIJ*)Y->data;
2426: PetscInt bs=Y->rmap->bs,bs2=bs*bs;
2427: PetscBLASInt one=1;
2430: if (str == SAME_NONZERO_PATTERN) {
2431: PetscScalar alpha = a;
2432: PetscBLASInt bnz;
2433: PetscBLASIntCast(x->nz*bs2,&bnz);
2434: PetscStackCallBLAS("BLASaxpy",BLASaxpy_(&bnz,&alpha,x->a,&one,y->a,&one));
2435: PetscObjectStateIncrease((PetscObject)Y);
2436: } else if (str == SUBSET_NONZERO_PATTERN) { /* nonzeros of X is a subset of Y's */
2437: MatAXPY_Basic(Y,a,X,str);
2438: } else {
2439: Mat B;
2440: PetscInt *nnz;
2441: if (bs != X->rmap->bs) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Matrices must have same block size");
2442: PetscMalloc1(Y->rmap->N,&nnz);
2443: MatCreate(PetscObjectComm((PetscObject)Y),&B);
2444: PetscObjectSetName((PetscObject)B,((PetscObject)Y)->name);
2445: MatSetSizes(B,Y->rmap->n,Y->cmap->n,Y->rmap->N,Y->cmap->N);
2446: MatSetBlockSizesFromMats(B,Y,Y);
2447: MatSetType(B,(MatType) ((PetscObject)Y)->type_name);
2448: MatAXPYGetPreallocation_SeqBAIJ(Y,X,nnz);
2449: MatSeqBAIJSetPreallocation(B,bs,0,nnz);
2450: MatAXPY_BasicWithPreallocation(B,Y,a,X,str);
2451: MatHeaderReplace(Y,&B);
2452: PetscFree(nnz);
2453: }
2454: return(0);
2455: }
2457: PetscErrorCode MatRealPart_SeqBAIJ(Mat A)
2458: {
2459: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
2460: PetscInt i,nz = a->bs2*a->i[a->mbs];
2461: MatScalar *aa = a->a;
2464: for (i=0; i<nz; i++) aa[i] = PetscRealPart(aa[i]);
2465: return(0);
2466: }
2468: PetscErrorCode MatImaginaryPart_SeqBAIJ(Mat A)
2469: {
2470: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
2471: PetscInt i,nz = a->bs2*a->i[a->mbs];
2472: MatScalar *aa = a->a;
2475: for (i=0; i<nz; i++) aa[i] = PetscImaginaryPart(aa[i]);
2476: return(0);
2477: }
2479: /*
2480: Code almost idential to MatGetColumnIJ_SeqAIJ() should share common code
2481: */
2482: PetscErrorCode MatGetColumnIJ_SeqBAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *nn,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done)
2483: {
2484: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
2486: PetscInt bs = A->rmap->bs,i,*collengths,*cia,*cja,n = A->cmap->n/bs,m = A->rmap->n/bs;
2487: PetscInt nz = a->i[m],row,*jj,mr,col;
2490: *nn = n;
2491: if (!ia) return(0);
2492: if (symmetric) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Not for BAIJ matrices");
2493: else {
2494: PetscCalloc1(n+1,&collengths);
2495: PetscMalloc1(n+1,&cia);
2496: PetscMalloc1(nz+1,&cja);
2497: jj = a->j;
2498: for (i=0; i<nz; i++) {
2499: collengths[jj[i]]++;
2500: }
2501: cia[0] = oshift;
2502: for (i=0; i<n; i++) {
2503: cia[i+1] = cia[i] + collengths[i];
2504: }
2505: PetscMemzero(collengths,n*sizeof(PetscInt));
2506: jj = a->j;
2507: for (row=0; row<m; row++) {
2508: mr = a->i[row+1] - a->i[row];
2509: for (i=0; i<mr; i++) {
2510: col = *jj++;
2512: cja[cia[col] + collengths[col]++ - oshift] = row + oshift;
2513: }
2514: }
2515: PetscFree(collengths);
2516: *ia = cia; *ja = cja;
2517: }
2518: return(0);
2519: }
2521: PetscErrorCode MatRestoreColumnIJ_SeqBAIJ(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscBool *done)
2522: {
2526: if (!ia) return(0);
2527: PetscFree(*ia);
2528: PetscFree(*ja);
2529: return(0);
2530: }
2532: /*
2533: MatGetColumnIJ_SeqBAIJ_Color() and MatRestoreColumnIJ_SeqBAIJ_Color() are customized from
2534: MatGetColumnIJ_SeqBAIJ() and MatRestoreColumnIJ_SeqBAIJ() by adding an output
2535: spidx[], index of a->a, to be used in MatTransposeColoringCreate() and MatFDColoringCreate()
2536: */
2537: PetscErrorCode MatGetColumnIJ_SeqBAIJ_Color(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *nn,const PetscInt *ia[],const PetscInt *ja[],PetscInt *spidx[],PetscBool *done)
2538: {
2539: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
2541: PetscInt i,*collengths,*cia,*cja,n=a->nbs,m=a->mbs;
2542: PetscInt nz = a->i[m],row,*jj,mr,col;
2543: PetscInt *cspidx;
2546: *nn = n;
2547: if (!ia) return(0);
2549: PetscCalloc1(n+1,&collengths);
2550: PetscMalloc1(n+1,&cia);
2551: PetscMalloc1(nz+1,&cja);
2552: PetscMalloc1(nz+1,&cspidx);
2553: jj = a->j;
2554: for (i=0; i<nz; i++) {
2555: collengths[jj[i]]++;
2556: }
2557: cia[0] = oshift;
2558: for (i=0; i<n; i++) {
2559: cia[i+1] = cia[i] + collengths[i];
2560: }
2561: PetscMemzero(collengths,n*sizeof(PetscInt));
2562: jj = a->j;
2563: for (row=0; row<m; row++) {
2564: mr = a->i[row+1] - a->i[row];
2565: for (i=0; i<mr; i++) {
2566: col = *jj++;
2567: cspidx[cia[col] + collengths[col] - oshift] = a->i[row] + i; /* index of a->j */
2568: cja[cia[col] + collengths[col]++ - oshift] = row + oshift;
2569: }
2570: }
2571: PetscFree(collengths);
2572: *ia = cia; *ja = cja;
2573: *spidx = cspidx;
2574: return(0);
2575: }
2577: PetscErrorCode MatRestoreColumnIJ_SeqBAIJ_Color(Mat A,PetscInt oshift,PetscBool symmetric,PetscBool inodecompressed,PetscInt *n,const PetscInt *ia[],const PetscInt *ja[],PetscInt *spidx[],PetscBool *done)
2578: {
2582: MatRestoreColumnIJ_SeqBAIJ(A,oshift,symmetric,inodecompressed,n,ia,ja,done);
2583: PetscFree(*spidx);
2584: return(0);
2585: }
2587: PetscErrorCode MatShift_SeqBAIJ(Mat Y,PetscScalar a)
2588: {
2590: Mat_SeqBAIJ *aij = (Mat_SeqBAIJ*)Y->data;
2593: if (!Y->preallocated || !aij->nz) {
2594: MatSeqBAIJSetPreallocation(Y,Y->rmap->bs,1,NULL);
2595: }
2596: MatShift_Basic(Y,a);
2597: return(0);
2598: }
2600: /* -------------------------------------------------------------------*/
2601: static struct _MatOps MatOps_Values = {MatSetValues_SeqBAIJ,
2602: MatGetRow_SeqBAIJ,
2603: MatRestoreRow_SeqBAIJ,
2604: MatMult_SeqBAIJ_N,
2605: /* 4*/ MatMultAdd_SeqBAIJ_N,
2606: MatMultTranspose_SeqBAIJ,
2607: MatMultTransposeAdd_SeqBAIJ,
2608: 0,
2609: 0,
2610: 0,
2611: /* 10*/ 0,
2612: MatLUFactor_SeqBAIJ,
2613: 0,
2614: 0,
2615: MatTranspose_SeqBAIJ,
2616: /* 15*/ MatGetInfo_SeqBAIJ,
2617: MatEqual_SeqBAIJ,
2618: MatGetDiagonal_SeqBAIJ,
2619: MatDiagonalScale_SeqBAIJ,
2620: MatNorm_SeqBAIJ,
2621: /* 20*/ 0,
2622: MatAssemblyEnd_SeqBAIJ,
2623: MatSetOption_SeqBAIJ,
2624: MatZeroEntries_SeqBAIJ,
2625: /* 24*/ MatZeroRows_SeqBAIJ,
2626: 0,
2627: 0,
2628: 0,
2629: 0,
2630: /* 29*/ MatSetUp_SeqBAIJ,
2631: 0,
2632: 0,
2633: 0,
2634: 0,
2635: /* 34*/ MatDuplicate_SeqBAIJ,
2636: 0,
2637: 0,
2638: MatILUFactor_SeqBAIJ,
2639: 0,
2640: /* 39*/ MatAXPY_SeqBAIJ,
2641: MatCreateSubMatrices_SeqBAIJ,
2642: MatIncreaseOverlap_SeqBAIJ,
2643: MatGetValues_SeqBAIJ,
2644: MatCopy_SeqBAIJ,
2645: /* 44*/ 0,
2646: MatScale_SeqBAIJ,
2647: MatShift_SeqBAIJ,
2648: 0,
2649: MatZeroRowsColumns_SeqBAIJ,
2650: /* 49*/ 0,
2651: MatGetRowIJ_SeqBAIJ,
2652: MatRestoreRowIJ_SeqBAIJ,
2653: MatGetColumnIJ_SeqBAIJ,
2654: MatRestoreColumnIJ_SeqBAIJ,
2655: /* 54*/ MatFDColoringCreate_SeqXAIJ,
2656: 0,
2657: 0,
2658: 0,
2659: MatSetValuesBlocked_SeqBAIJ,
2660: /* 59*/ MatCreateSubMatrix_SeqBAIJ,
2661: MatDestroy_SeqBAIJ,
2662: MatView_SeqBAIJ,
2663: 0,
2664: 0,
2665: /* 64*/ 0,
2666: 0,
2667: 0,
2668: 0,
2669: 0,
2670: /* 69*/ MatGetRowMaxAbs_SeqBAIJ,
2671: 0,
2672: MatConvert_Basic,
2673: 0,
2674: 0,
2675: /* 74*/ 0,
2676: MatFDColoringApply_BAIJ,
2677: 0,
2678: 0,
2679: 0,
2680: /* 79*/ 0,
2681: 0,
2682: 0,
2683: 0,
2684: MatLoad_SeqBAIJ,
2685: /* 84*/ 0,
2686: 0,
2687: 0,
2688: 0,
2689: 0,
2690: /* 89*/ 0,
2691: 0,
2692: 0,
2693: 0,
2694: 0,
2695: /* 94*/ 0,
2696: 0,
2697: 0,
2698: 0,
2699: 0,
2700: /* 99*/ 0,
2701: 0,
2702: 0,
2703: 0,
2704: 0,
2705: /*104*/ 0,
2706: MatRealPart_SeqBAIJ,
2707: MatImaginaryPart_SeqBAIJ,
2708: 0,
2709: 0,
2710: /*109*/ 0,
2711: 0,
2712: 0,
2713: 0,
2714: MatMissingDiagonal_SeqBAIJ,
2715: /*114*/ 0,
2716: 0,
2717: 0,
2718: 0,
2719: 0,
2720: /*119*/ 0,
2721: 0,
2722: MatMultHermitianTranspose_SeqBAIJ,
2723: MatMultHermitianTransposeAdd_SeqBAIJ,
2724: 0,
2725: /*124*/ 0,
2726: 0,
2727: MatInvertBlockDiagonal_SeqBAIJ,
2728: 0,
2729: 0,
2730: /*129*/ 0,
2731: 0,
2732: 0,
2733: 0,
2734: 0,
2735: /*134*/ 0,
2736: 0,
2737: 0,
2738: 0,
2739: 0,
2740: /*139*/ MatSetBlockSizes_Default,
2741: 0,
2742: 0,
2743: MatFDColoringSetUp_SeqXAIJ,
2744: 0,
2745: /*144*/MatCreateMPIMatConcatenateSeqMat_SeqBAIJ,
2746: MatDestroySubMatrices_SeqBAIJ
2747: };
2749: PetscErrorCode MatStoreValues_SeqBAIJ(Mat mat)
2750: {
2751: Mat_SeqBAIJ *aij = (Mat_SeqBAIJ*)mat->data;
2752: PetscInt nz = aij->i[aij->mbs]*aij->bs2;
2756: if (aij->nonew != 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first");
2758: /* allocate space for values if not already there */
2759: if (!aij->saved_values) {
2760: PetscMalloc1(nz+1,&aij->saved_values);
2761: PetscLogObjectMemory((PetscObject)mat,(nz+1)*sizeof(PetscScalar));
2762: }
2764: /* copy values over */
2765: PetscMemcpy(aij->saved_values,aij->a,nz*sizeof(PetscScalar));
2766: return(0);
2767: }
2769: PetscErrorCode MatRetrieveValues_SeqBAIJ(Mat mat)
2770: {
2771: Mat_SeqBAIJ *aij = (Mat_SeqBAIJ*)mat->data;
2773: PetscInt nz = aij->i[aij->mbs]*aij->bs2;
2776: if (aij->nonew != 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first");
2777: if (!aij->saved_values) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Must call MatStoreValues(A);first");
2779: /* copy values over */
2780: PetscMemcpy(aij->a,aij->saved_values,nz*sizeof(PetscScalar));
2781: return(0);
2782: }
2784: PETSC_INTERN PetscErrorCode MatConvert_SeqBAIJ_SeqAIJ(Mat, MatType,MatReuse,Mat*);
2785: PETSC_INTERN PetscErrorCode MatConvert_SeqBAIJ_SeqSBAIJ(Mat, MatType,MatReuse,Mat*);
2787: PetscErrorCode MatSeqBAIJSetPreallocation_SeqBAIJ(Mat B,PetscInt bs,PetscInt nz,PetscInt *nnz)
2788: {
2789: Mat_SeqBAIJ *b;
2791: PetscInt i,mbs,nbs,bs2;
2792: PetscBool flg = PETSC_FALSE,skipallocation = PETSC_FALSE,realalloc = PETSC_FALSE;
2795: if (nz >= 0 || nnz) realalloc = PETSC_TRUE;
2796: if (nz == MAT_SKIP_ALLOCATION) {
2797: skipallocation = PETSC_TRUE;
2798: nz = 0;
2799: }
2801: MatSetBlockSize(B,PetscAbs(bs));
2802: PetscLayoutSetUp(B->rmap);
2803: PetscLayoutSetUp(B->cmap);
2804: PetscLayoutGetBlockSize(B->rmap,&bs);
2806: B->preallocated = PETSC_TRUE;
2808: mbs = B->rmap->n/bs;
2809: nbs = B->cmap->n/bs;
2810: bs2 = bs*bs;
2812: if (mbs*bs!=B->rmap->n || nbs*bs!=B->cmap->n) SETERRQ3(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Number rows %D, cols %D must be divisible by blocksize %D",B->rmap->N,B->cmap->n,bs);
2814: if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 5;
2815: if (nz < 0) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"nz cannot be less than 0: value %D",nz);
2816: if (nnz) {
2817: for (i=0; i<mbs; i++) {
2818: if (nnz[i] < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"nnz cannot be less than 0: local row %D value %D",i,nnz[i]);
2819: if (nnz[i] > nbs) SETERRQ3(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"nnz cannot be greater than block row length: local row %D value %D rowlength %D",i,nnz[i],nbs);
2820: }
2821: }
2823: b = (Mat_SeqBAIJ*)B->data;
2824: PetscOptionsBegin(PetscObjectComm((PetscObject)B),NULL,"Optimize options for SEQBAIJ matrix 2 ","Mat");
2825: PetscOptionsBool("-mat_no_unroll","Do not optimize for block size (slow)",NULL,flg,&flg,NULL);
2826: PetscOptionsEnd();
2828: if (!flg) {
2829: switch (bs) {
2830: case 1:
2831: B->ops->mult = MatMult_SeqBAIJ_1;
2832: B->ops->multadd = MatMultAdd_SeqBAIJ_1;
2833: break;
2834: case 2:
2835: B->ops->mult = MatMult_SeqBAIJ_2;
2836: B->ops->multadd = MatMultAdd_SeqBAIJ_2;
2837: break;
2838: case 3:
2839: B->ops->mult = MatMult_SeqBAIJ_3;
2840: B->ops->multadd = MatMultAdd_SeqBAIJ_3;
2841: break;
2842: case 4:
2843: B->ops->mult = MatMult_SeqBAIJ_4;
2844: B->ops->multadd = MatMultAdd_SeqBAIJ_4;
2845: break;
2846: case 5:
2847: B->ops->mult = MatMult_SeqBAIJ_5;
2848: B->ops->multadd = MatMultAdd_SeqBAIJ_5;
2849: break;
2850: case 6:
2851: B->ops->mult = MatMult_SeqBAIJ_6;
2852: B->ops->multadd = MatMultAdd_SeqBAIJ_6;
2853: break;
2854: case 7:
2855: B->ops->mult = MatMult_SeqBAIJ_7;
2856: B->ops->multadd = MatMultAdd_SeqBAIJ_7;
2857: break;
2858: case 9:
2859: #if defined(PETSC_HAVE_IMMINTRIN_H) && defined(__AVX2__) && defined(__FMA__) && defined(PETSC_USE_REAL_DOUBLE) && !defined(PETSC_USE_COMPLEX) && !defined(PETSC_USE_64BIT_INDICES)
2860: B->ops->mult = MatMult_SeqBAIJ_9_AVX2;
2861: B->ops->multadd = MatMultAdd_SeqBAIJ_9_AVX2;
2862: #else
2863: B->ops->mult = MatMult_SeqBAIJ_N;
2864: B->ops->multadd = MatMultAdd_SeqBAIJ_N;
2865: #endif
2866: break;
2867: case 11:
2868: B->ops->mult = MatMult_SeqBAIJ_11;
2869: B->ops->multadd = MatMultAdd_SeqBAIJ_11;
2870: break;
2871: case 15:
2872: B->ops->mult = MatMult_SeqBAIJ_15_ver1;
2873: B->ops->multadd = MatMultAdd_SeqBAIJ_N;
2874: break;
2875: default:
2876: B->ops->mult = MatMult_SeqBAIJ_N;
2877: B->ops->multadd = MatMultAdd_SeqBAIJ_N;
2878: break;
2879: }
2880: }
2881: B->ops->sor = MatSOR_SeqBAIJ;
2882: b->mbs = mbs;
2883: b->nbs = nbs;
2884: if (!skipallocation) {
2885: if (!b->imax) {
2886: PetscMalloc2(mbs,&b->imax,mbs,&b->ilen);
2887: PetscLogObjectMemory((PetscObject)B,2*mbs*sizeof(PetscInt));
2889: b->free_imax_ilen = PETSC_TRUE;
2890: }
2891: /* b->ilen will count nonzeros in each block row so far. */
2892: for (i=0; i<mbs; i++) b->ilen[i] = 0;
2893: if (!nnz) {
2894: if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 5;
2895: else if (nz < 0) nz = 1;
2896: for (i=0; i<mbs; i++) b->imax[i] = nz;
2897: nz = nz*mbs;
2898: } else {
2899: nz = 0;
2900: for (i=0; i<mbs; i++) {b->imax[i] = nnz[i]; nz += nnz[i];}
2901: }
2903: /* allocate the matrix space */
2904: MatSeqXAIJFreeAIJ(B,&b->a,&b->j,&b->i);
2905: if (B->structure_only) {
2906: PetscMalloc1(nz,&b->j);
2907: PetscMalloc1(B->rmap->N+1,&b->i);
2908: PetscLogObjectMemory((PetscObject)B,(B->rmap->N+1)*sizeof(PetscInt)+nz*sizeof(PetscInt));
2909: } else {
2910: PetscMalloc3(bs2*nz,&b->a,nz,&b->j,B->rmap->N+1,&b->i);
2911: PetscLogObjectMemory((PetscObject)B,(B->rmap->N+1)*sizeof(PetscInt)+nz*(bs2*sizeof(PetscScalar)+sizeof(PetscInt)));
2912: PetscMemzero(b->a,nz*bs2*sizeof(MatScalar));
2913: }
2914: PetscMemzero(b->j,nz*sizeof(PetscInt));
2916: if (B->structure_only) {
2917: b->singlemalloc = PETSC_FALSE;
2918: b->free_a = PETSC_FALSE;
2919: } else {
2920: b->singlemalloc = PETSC_TRUE;
2921: b->free_a = PETSC_TRUE;
2922: }
2923: b->free_ij = PETSC_TRUE;
2925: b->i[0] = 0;
2926: for (i=1; i<mbs+1; i++) {
2927: b->i[i] = b->i[i-1] + b->imax[i-1];
2928: }
2930: } else {
2931: b->free_a = PETSC_FALSE;
2932: b->free_ij = PETSC_FALSE;
2933: }
2935: b->bs2 = bs2;
2936: b->mbs = mbs;
2937: b->nz = 0;
2938: b->maxnz = nz;
2939: B->info.nz_unneeded = (PetscReal)b->maxnz*bs2;
2940: B->was_assembled = PETSC_FALSE;
2941: B->assembled = PETSC_FALSE;
2942: if (realalloc) {MatSetOption(B,MAT_NEW_NONZERO_ALLOCATION_ERR,PETSC_TRUE);}
2943: return(0);
2944: }
2946: PetscErrorCode MatSeqBAIJSetPreallocationCSR_SeqBAIJ(Mat B,PetscInt bs,const PetscInt ii[],const PetscInt jj[],const PetscScalar V[])
2947: {
2948: PetscInt i,m,nz,nz_max=0,*nnz;
2949: PetscScalar *values=0;
2950: PetscBool roworiented = ((Mat_SeqBAIJ*)B->data)->roworiented;
2954: if (bs < 1) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Invalid block size specified, must be positive but it is %D",bs);
2955: PetscLayoutSetBlockSize(B->rmap,bs);
2956: PetscLayoutSetBlockSize(B->cmap,bs);
2957: PetscLayoutSetUp(B->rmap);
2958: PetscLayoutSetUp(B->cmap);
2959: PetscLayoutGetBlockSize(B->rmap,&bs);
2960: m = B->rmap->n/bs;
2962: if (ii[0] != 0) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE, "ii[0] must be 0 but it is %D",ii[0]);
2963: PetscMalloc1(m+1, &nnz);
2964: for (i=0; i<m; i++) {
2965: nz = ii[i+1]- ii[i];
2966: if (nz < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE, "Local row %D has a negative number of columns %D",i,nz);
2967: nz_max = PetscMax(nz_max, nz);
2968: nnz[i] = nz;
2969: }
2970: MatSeqBAIJSetPreallocation(B,bs,0,nnz);
2971: PetscFree(nnz);
2973: values = (PetscScalar*)V;
2974: if (!values) {
2975: PetscCalloc1(bs*bs*(nz_max+1),&values);
2976: }
2977: for (i=0; i<m; i++) {
2978: PetscInt ncols = ii[i+1] - ii[i];
2979: const PetscInt *icols = jj + ii[i];
2980: const PetscScalar *svals = values + (V ? (bs*bs*ii[i]) : 0);
2981: if (!roworiented) {
2982: MatSetValuesBlocked_SeqBAIJ(B,1,&i,ncols,icols,svals,INSERT_VALUES);
2983: } else {
2984: PetscInt j;
2985: for (j=0; j<ncols; j++) {
2986: const PetscScalar *svals = values + (V ? (bs*bs*(ii[i]+j)) : 0);
2987: MatSetValuesBlocked_SeqBAIJ(B,1,&i,1,&icols[j],svals,INSERT_VALUES);
2988: }
2989: }
2990: }
2991: if (!V) { PetscFree(values); }
2992: MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);
2993: MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);
2994: MatSetOption(B,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_TRUE);
2995: return(0);
2996: }
2998: /*MC
2999: MATSEQBAIJ - MATSEQBAIJ = "seqbaij" - A matrix type to be used for sequential block sparse matrices, based on
3000: block sparse compressed row format.
3002: Options Database Keys:
3003: . -mat_type seqbaij - sets the matrix type to "seqbaij" during a call to MatSetFromOptions()
3005: Level: beginner
3007: .seealso: MatCreateSeqBAIJ()
3008: M*/
3010: PETSC_INTERN PetscErrorCode MatConvert_SeqBAIJ_SeqBSTRM(Mat, MatType,MatReuse,Mat*);
3012: PETSC_EXTERN PetscErrorCode MatCreate_SeqBAIJ(Mat B)
3013: {
3015: PetscMPIInt size;
3016: Mat_SeqBAIJ *b;
3019: MPI_Comm_size(PetscObjectComm((PetscObject)B),&size);
3020: if (size > 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Comm must be of size 1");
3022: PetscNewLog(B,&b);
3023: B->data = (void*)b;
3024: PetscMemcpy(B->ops,&MatOps_Values,sizeof(struct _MatOps));
3026: b->row = 0;
3027: b->col = 0;
3028: b->icol = 0;
3029: b->reallocs = 0;
3030: b->saved_values = 0;
3032: b->roworiented = PETSC_TRUE;
3033: b->nonew = 0;
3034: b->diag = 0;
3035: B->spptr = 0;
3036: B->info.nz_unneeded = (PetscReal)b->maxnz*b->bs2;
3037: b->keepnonzeropattern = PETSC_FALSE;
3039: PetscObjectComposeFunction((PetscObject)B,"MatInvertBlockDiagonal_C",MatInvertBlockDiagonal_SeqBAIJ);
3040: PetscObjectComposeFunction((PetscObject)B,"MatStoreValues_C",MatStoreValues_SeqBAIJ);
3041: PetscObjectComposeFunction((PetscObject)B,"MatRetrieveValues_C",MatRetrieveValues_SeqBAIJ);
3042: PetscObjectComposeFunction((PetscObject)B,"MatSeqBAIJSetColumnIndices_C",MatSeqBAIJSetColumnIndices_SeqBAIJ);
3043: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqbaij_seqaij_C",MatConvert_SeqBAIJ_SeqAIJ);
3044: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqbaij_seqsbaij_C",MatConvert_SeqBAIJ_SeqSBAIJ);
3045: PetscObjectComposeFunction((PetscObject)B,"MatSeqBAIJSetPreallocation_C",MatSeqBAIJSetPreallocation_SeqBAIJ);
3046: PetscObjectComposeFunction((PetscObject)B,"MatSeqBAIJSetPreallocationCSR_C",MatSeqBAIJSetPreallocationCSR_SeqBAIJ);
3047: PetscObjectComposeFunction((PetscObject)B,"MatIsTranspose_C",MatIsTranspose_SeqBAIJ);
3048: #if defined(PETSC_HAVE_HYPRE)
3049: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqbaij_hypre_C",MatConvert_AIJ_HYPRE);
3050: #endif
3051: PetscObjectComposeFunction((PetscObject)B,"MatConvert_seqbaij_is_C",MatConvert_XAIJ_IS);
3052: PetscObjectComposeFunction((PetscObject)B,"MatPtAP_is_seqbaij_C",MatPtAP_IS_XAIJ);
3053: PetscObjectChangeTypeName((PetscObject)B,MATSEQBAIJ);
3054: return(0);
3055: }
3057: PetscErrorCode MatDuplicateNoCreate_SeqBAIJ(Mat C,Mat A,MatDuplicateOption cpvalues,PetscBool mallocmatspace)
3058: {
3059: Mat_SeqBAIJ *c = (Mat_SeqBAIJ*)C->data,*a = (Mat_SeqBAIJ*)A->data;
3061: PetscInt i,mbs = a->mbs,nz = a->nz,bs2 = a->bs2;
3064: if (a->i[mbs] != nz) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Corrupt matrix");
3066: if (cpvalues == MAT_SHARE_NONZERO_PATTERN) {
3067: c->imax = a->imax;
3068: c->ilen = a->ilen;
3069: c->free_imax_ilen = PETSC_FALSE;
3070: } else {
3071: PetscMalloc2(mbs,&c->imax,mbs,&c->ilen);
3072: PetscLogObjectMemory((PetscObject)C,2*mbs*sizeof(PetscInt));
3073: for (i=0; i<mbs; i++) {
3074: c->imax[i] = a->imax[i];
3075: c->ilen[i] = a->ilen[i];
3076: }
3077: c->free_imax_ilen = PETSC_TRUE;
3078: }
3080: /* allocate the matrix space */
3081: if (mallocmatspace) {
3082: if (cpvalues == MAT_SHARE_NONZERO_PATTERN) {
3083: PetscCalloc1(bs2*nz,&c->a);
3084: PetscLogObjectMemory((PetscObject)C,a->i[mbs]*bs2*sizeof(PetscScalar));
3086: c->i = a->i;
3087: c->j = a->j;
3088: c->singlemalloc = PETSC_FALSE;
3089: c->free_a = PETSC_TRUE;
3090: c->free_ij = PETSC_FALSE;
3091: c->parent = A;
3092: C->preallocated = PETSC_TRUE;
3093: C->assembled = PETSC_TRUE;
3095: PetscObjectReference((PetscObject)A);
3096: MatSetOption(A,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_TRUE);
3097: MatSetOption(C,MAT_NEW_NONZERO_LOCATION_ERR,PETSC_TRUE);
3098: } else {
3099: PetscMalloc3(bs2*nz,&c->a,nz,&c->j,mbs+1,&c->i);
3100: PetscLogObjectMemory((PetscObject)C,a->i[mbs]*(bs2*sizeof(PetscScalar)+sizeof(PetscInt))+(mbs+1)*sizeof(PetscInt));
3102: c->singlemalloc = PETSC_TRUE;
3103: c->free_a = PETSC_TRUE;
3104: c->free_ij = PETSC_TRUE;
3106: PetscMemcpy(c->i,a->i,(mbs+1)*sizeof(PetscInt));
3107: if (mbs > 0) {
3108: PetscMemcpy(c->j,a->j,nz*sizeof(PetscInt));
3109: if (cpvalues == MAT_COPY_VALUES) {
3110: PetscMemcpy(c->a,a->a,bs2*nz*sizeof(MatScalar));
3111: } else {
3112: PetscMemzero(c->a,bs2*nz*sizeof(MatScalar));
3113: }
3114: }
3115: C->preallocated = PETSC_TRUE;
3116: C->assembled = PETSC_TRUE;
3117: }
3118: }
3120: c->roworiented = a->roworiented;
3121: c->nonew = a->nonew;
3123: PetscLayoutReference(A->rmap,&C->rmap);
3124: PetscLayoutReference(A->cmap,&C->cmap);
3126: c->bs2 = a->bs2;
3127: c->mbs = a->mbs;
3128: c->nbs = a->nbs;
3130: if (a->diag) {
3131: if (cpvalues == MAT_SHARE_NONZERO_PATTERN) {
3132: c->diag = a->diag;
3133: c->free_diag = PETSC_FALSE;
3134: } else {
3135: PetscMalloc1(mbs+1,&c->diag);
3136: PetscLogObjectMemory((PetscObject)C,(mbs+1)*sizeof(PetscInt));
3137: for (i=0; i<mbs; i++) c->diag[i] = a->diag[i];
3138: c->free_diag = PETSC_TRUE;
3139: }
3140: } else c->diag = 0;
3142: c->nz = a->nz;
3143: c->maxnz = a->nz; /* Since we allocate exactly the right amount */
3144: c->solve_work = NULL;
3145: c->mult_work = NULL;
3146: c->sor_workt = NULL;
3147: c->sor_work = NULL;
3149: c->compressedrow.use = a->compressedrow.use;
3150: c->compressedrow.nrows = a->compressedrow.nrows;
3151: if (a->compressedrow.use) {
3152: i = a->compressedrow.nrows;
3153: PetscMalloc2(i+1,&c->compressedrow.i,i+1,&c->compressedrow.rindex);
3154: PetscLogObjectMemory((PetscObject)C,(2*i+1)*sizeof(PetscInt));
3155: PetscMemcpy(c->compressedrow.i,a->compressedrow.i,(i+1)*sizeof(PetscInt));
3156: PetscMemcpy(c->compressedrow.rindex,a->compressedrow.rindex,i*sizeof(PetscInt));
3157: } else {
3158: c->compressedrow.use = PETSC_FALSE;
3159: c->compressedrow.i = NULL;
3160: c->compressedrow.rindex = NULL;
3161: }
3162: C->nonzerostate = A->nonzerostate;
3164: PetscFunctionListDuplicate(((PetscObject)A)->qlist,&((PetscObject)C)->qlist);
3165: return(0);
3166: }
3168: PetscErrorCode MatDuplicate_SeqBAIJ(Mat A,MatDuplicateOption cpvalues,Mat *B)
3169: {
3173: MatCreate(PetscObjectComm((PetscObject)A),B);
3174: MatSetSizes(*B,A->rmap->N,A->cmap->n,A->rmap->N,A->cmap->n);
3175: MatSetType(*B,MATSEQBAIJ);
3176: MatDuplicateNoCreate_SeqBAIJ(*B,A,cpvalues,PETSC_TRUE);
3177: return(0);
3178: }
3180: PetscErrorCode MatLoad_SeqBAIJ(Mat newmat,PetscViewer viewer)
3181: {
3182: Mat_SeqBAIJ *a;
3184: PetscInt i,nz,header[4],*rowlengths=0,M,N,bs = newmat->rmap->bs;
3185: PetscInt *mask,mbs,*jj,j,rowcount,nzcount,k,*browlengths,maskcount;
3186: PetscInt kmax,jcount,block,idx,point,nzcountb,extra_rows,rows,cols;
3187: PetscInt *masked,nmask,tmp,bs2,ishift;
3188: PetscMPIInt size;
3189: int fd;
3190: PetscScalar *aa;
3191: MPI_Comm comm;
3194: /* force binary viewer to load .info file if it has not yet done so */
3195: PetscViewerSetUp(viewer);
3196: PetscObjectGetComm((PetscObject)viewer,&comm);
3197: PetscOptionsBegin(comm,NULL,"Options for loading SEQBAIJ matrix","Mat");
3198: PetscOptionsInt("-matload_block_size","Set the blocksize used to store the matrix","MatLoad",bs,&bs,NULL);
3199: PetscOptionsEnd();
3200: if (bs < 0) bs = 1;
3201: bs2 = bs*bs;
3203: MPI_Comm_size(comm,&size);
3204: if (size > 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"view must have one processor");
3205: PetscViewerBinaryGetDescriptor(viewer,&fd);
3206: PetscBinaryRead(fd,header,4,PETSC_INT);
3207: if (header[0] != MAT_FILE_CLASSID) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"not Mat object");
3208: M = header[1]; N = header[2]; nz = header[3];
3210: if (header[3] < 0) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Matrix stored in special format, cannot load as SeqBAIJ");
3211: if (M != N) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Can only do square matrices");
3213: /*
3214: This code adds extra rows to make sure the number of rows is
3215: divisible by the blocksize
3216: */
3217: mbs = M/bs;
3218: extra_rows = bs - M + bs*(mbs);
3219: if (extra_rows == bs) extra_rows = 0;
3220: else mbs++;
3221: if (extra_rows) {
3222: PetscInfo(viewer,"Padding loaded matrix to match blocksize\n");
3223: }
3225: /* Set global sizes if not already set */
3226: if (newmat->rmap->n < 0 && newmat->rmap->N < 0 && newmat->cmap->n < 0 && newmat->cmap->N < 0) {
3227: MatSetSizes(newmat,PETSC_DECIDE,PETSC_DECIDE,M+extra_rows,N+extra_rows);
3228: } else { /* Check if the matrix global sizes are correct */
3229: MatGetSize(newmat,&rows,&cols);
3230: if (rows < 0 && cols < 0) { /* user might provide local size instead of global size */
3231: MatGetLocalSize(newmat,&rows,&cols);
3232: }
3233: if (M != rows || N != cols) SETERRQ4(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Matrix in file of different length (%d, %d) than the input matrix (%d, %d)",M,N,rows,cols);
3234: }
3236: /* read in row lengths */
3237: PetscMalloc1(M+extra_rows,&rowlengths);
3238: PetscBinaryRead(fd,rowlengths,M,PETSC_INT);
3239: for (i=0; i<extra_rows; i++) rowlengths[M+i] = 1;
3241: /* read in column indices */
3242: PetscMalloc1(nz+extra_rows,&jj);
3243: PetscBinaryRead(fd,jj,nz,PETSC_INT);
3244: for (i=0; i<extra_rows; i++) jj[nz+i] = M+i;
3246: /* loop over row lengths determining block row lengths */
3247: PetscCalloc1(mbs,&browlengths);
3248: PetscMalloc2(mbs,&mask,mbs,&masked);
3249: PetscMemzero(mask,mbs*sizeof(PetscInt));
3250: rowcount = 0;
3251: nzcount = 0;
3252: for (i=0; i<mbs; i++) {
3253: nmask = 0;
3254: for (j=0; j<bs; j++) {
3255: kmax = rowlengths[rowcount];
3256: for (k=0; k<kmax; k++) {
3257: tmp = jj[nzcount++]/bs;
3258: if (!mask[tmp]) {masked[nmask++] = tmp; mask[tmp] = 1;}
3259: }
3260: rowcount++;
3261: }
3262: browlengths[i] += nmask;
3263: /* zero out the mask elements we set */
3264: for (j=0; j<nmask; j++) mask[masked[j]] = 0;
3265: }
3267: /* Do preallocation */
3268: MatSeqBAIJSetPreallocation(newmat,bs,0,browlengths);
3269: a = (Mat_SeqBAIJ*)newmat->data;
3271: /* set matrix "i" values */
3272: a->i[0] = 0;
3273: for (i=1; i<= mbs; i++) {
3274: a->i[i] = a->i[i-1] + browlengths[i-1];
3275: a->ilen[i-1] = browlengths[i-1];
3276: }
3277: a->nz = 0;
3278: for (i=0; i<mbs; i++) a->nz += browlengths[i];
3280: /* read in nonzero values */
3281: PetscMalloc1(nz+extra_rows,&aa);
3282: PetscBinaryRead(fd,aa,nz,PETSC_SCALAR);
3283: for (i=0; i<extra_rows; i++) aa[nz+i] = 1.0;
3285: /* set "a" and "j" values into matrix */
3286: nzcount = 0; jcount = 0;
3287: for (i=0; i<mbs; i++) {
3288: nzcountb = nzcount;
3289: nmask = 0;
3290: for (j=0; j<bs; j++) {
3291: kmax = rowlengths[i*bs+j];
3292: for (k=0; k<kmax; k++) {
3293: tmp = jj[nzcount++]/bs;
3294: if (!mask[tmp]) { masked[nmask++] = tmp; mask[tmp] = 1;}
3295: }
3296: }
3297: /* sort the masked values */
3298: PetscSortInt(nmask,masked);
3300: /* set "j" values into matrix */
3301: maskcount = 1;
3302: for (j=0; j<nmask; j++) {
3303: a->j[jcount++] = masked[j];
3304: mask[masked[j]] = maskcount++;
3305: }
3306: /* set "a" values into matrix */
3307: ishift = bs2*a->i[i];
3308: for (j=0; j<bs; j++) {
3309: kmax = rowlengths[i*bs+j];
3310: for (k=0; k<kmax; k++) {
3311: tmp = jj[nzcountb]/bs;
3312: block = mask[tmp] - 1;
3313: point = jj[nzcountb] - bs*tmp;
3314: idx = ishift + bs2*block + j + bs*point;
3315: a->a[idx] = (MatScalar)aa[nzcountb++];
3316: }
3317: }
3318: /* zero out the mask elements we set */
3319: for (j=0; j<nmask; j++) mask[masked[j]] = 0;
3320: }
3321: if (jcount != a->nz) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_FILE_UNEXPECTED,"Bad binary matrix");
3323: PetscFree(rowlengths);
3324: PetscFree(browlengths);
3325: PetscFree(aa);
3326: PetscFree(jj);
3327: PetscFree2(mask,masked);
3329: MatAssemblyBegin(newmat,MAT_FINAL_ASSEMBLY);
3330: MatAssemblyEnd(newmat,MAT_FINAL_ASSEMBLY);
3331: return(0);
3332: }
3334: /*@C
3335: MatCreateSeqBAIJ - Creates a sparse matrix in block AIJ (block
3336: compressed row) format. For good matrix assembly performance the
3337: user should preallocate the matrix storage by setting the parameter nz
3338: (or the array nnz). By setting these parameters accurately, performance
3339: during matrix assembly can be increased by more than a factor of 50.
3341: Collective on MPI_Comm
3343: Input Parameters:
3344: + comm - MPI communicator, set to PETSC_COMM_SELF
3345: . bs - size of block, the blocks are ALWAYS square. One can use MatSetBlockSizes() to set a different row and column blocksize but the row
3346: blocksize always defines the size of the blocks. The column blocksize sets the blocksize of the vectors obtained with MatCreateVecs()
3347: . m - number of rows
3348: . n - number of columns
3349: . nz - number of nonzero blocks per block row (same for all rows)
3350: - nnz - array containing the number of nonzero blocks in the various block rows
3351: (possibly different for each block row) or NULL
3353: Output Parameter:
3354: . A - the matrix
3356: It is recommended that one use the MatCreate(), MatSetType() and/or MatSetFromOptions(),
3357: MatXXXXSetPreallocation() paradgm instead of this routine directly.
3358: [MatXXXXSetPreallocation() is, for example, MatSeqAIJSetPreallocation]
3360: Options Database Keys:
3361: . -mat_no_unroll - uses code that does not unroll the loops in the
3362: block calculations (much slower)
3363: . -mat_block_size - size of the blocks to use
3365: Level: intermediate
3367: Notes:
3368: The number of rows and columns must be divisible by blocksize.
3370: If the nnz parameter is given then the nz parameter is ignored
3372: A nonzero block is any block that as 1 or more nonzeros in it
3374: The block AIJ format is fully compatible with standard Fortran 77
3375: storage. That is, the stored row and column indices can begin at
3376: either one (as in Fortran) or zero. See the users' manual for details.
3378: Specify the preallocated storage with either nz or nnz (not both).
3379: Set nz=PETSC_DEFAULT and nnz=NULL for PETSc to control dynamic memory
3380: allocation. See Users-Manual: ch_mat for details.
3381: matrices.
3383: .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatCreateBAIJ()
3384: @*/
3385: PetscErrorCode MatCreateSeqBAIJ(MPI_Comm comm,PetscInt bs,PetscInt m,PetscInt n,PetscInt nz,const PetscInt nnz[],Mat *A)
3386: {
3390: MatCreate(comm,A);
3391: MatSetSizes(*A,m,n,m,n);
3392: MatSetType(*A,MATSEQBAIJ);
3393: MatSeqBAIJSetPreallocation(*A,bs,nz,(PetscInt*)nnz);
3394: return(0);
3395: }
3397: /*@C
3398: MatSeqBAIJSetPreallocation - Sets the block size and expected nonzeros
3399: per row in the matrix. For good matrix assembly performance the
3400: user should preallocate the matrix storage by setting the parameter nz
3401: (or the array nnz). By setting these parameters accurately, performance
3402: during matrix assembly can be increased by more than a factor of 50.
3404: Collective on MPI_Comm
3406: Input Parameters:
3407: + B - the matrix
3408: . bs - size of block, the blocks are ALWAYS square. One can use MatSetBlockSizes() to set a different row and column blocksize but the row
3409: blocksize always defines the size of the blocks. The column blocksize sets the blocksize of the vectors obtained with MatCreateVecs()
3410: . nz - number of block nonzeros per block row (same for all rows)
3411: - nnz - array containing the number of block nonzeros in the various block rows
3412: (possibly different for each block row) or NULL
3414: Options Database Keys:
3415: . -mat_no_unroll - uses code that does not unroll the loops in the
3416: block calculations (much slower)
3417: . -mat_block_size - size of the blocks to use
3419: Level: intermediate
3421: Notes:
3422: If the nnz parameter is given then the nz parameter is ignored
3424: You can call MatGetInfo() to get information on how effective the preallocation was;
3425: for example the fields mallocs,nz_allocated,nz_used,nz_unneeded;
3426: You can also run with the option -info and look for messages with the string
3427: malloc in them to see if additional memory allocation was needed.
3429: The block AIJ format is fully compatible with standard Fortran 77
3430: storage. That is, the stored row and column indices can begin at
3431: either one (as in Fortran) or zero. See the users' manual for details.
3433: Specify the preallocated storage with either nz or nnz (not both).
3434: Set nz=PETSC_DEFAULT and nnz=NULL for PETSc to control dynamic memory
3435: allocation. See Users-Manual: ch_mat for details.
3437: .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatCreateBAIJ(), MatGetInfo()
3438: @*/
3439: PetscErrorCode MatSeqBAIJSetPreallocation(Mat B,PetscInt bs,PetscInt nz,const PetscInt nnz[])
3440: {
3447: PetscTryMethod(B,"MatSeqBAIJSetPreallocation_C",(Mat,PetscInt,PetscInt,const PetscInt[]),(B,bs,nz,nnz));
3448: return(0);
3449: }
3451: /*@C
3452: MatSeqBAIJSetPreallocationCSR - Allocates memory for a sparse sequential matrix in AIJ format
3453: (the default sequential PETSc format).
3455: Collective on MPI_Comm
3457: Input Parameters:
3458: + B - the matrix
3459: . i - the indices into j for the start of each local row (starts with zero)
3460: . j - the column indices for each local row (starts with zero) these must be sorted for each row
3461: - v - optional values in the matrix
3463: Level: developer
3465: Notes:
3466: The order of the entries in values is specified by the MatOption MAT_ROW_ORIENTED. For example, C programs
3467: may want to use the default MAT_ROW_ORIENTED=PETSC_TRUE and use an array v[nnz][bs][bs] where the second index is
3468: over rows within a block and the last index is over columns within a block row. Fortran programs will likely set
3469: MAT_ROW_ORIENTED=PETSC_FALSE and use a Fortran array v(bs,bs,nnz) in which the first index is over rows within a
3470: block column and the second index is over columns within a block.
3472: .keywords: matrix, aij, compressed row, sparse
3474: .seealso: MatCreate(), MatCreateSeqBAIJ(), MatSetValues(), MatSeqBAIJSetPreallocation(), MATSEQBAIJ
3475: @*/
3476: PetscErrorCode MatSeqBAIJSetPreallocationCSR(Mat B,PetscInt bs,const PetscInt i[],const PetscInt j[], const PetscScalar v[])
3477: {
3484: PetscTryMethod(B,"MatSeqBAIJSetPreallocationCSR_C",(Mat,PetscInt,const PetscInt[],const PetscInt[],const PetscScalar[]),(B,bs,i,j,v));
3485: return(0);
3486: }
3489: /*@
3490: MatCreateSeqBAIJWithArrays - Creates an sequential BAIJ matrix using matrix elements provided by the user.
3492: Collective on MPI_Comm
3494: Input Parameters:
3495: + comm - must be an MPI communicator of size 1
3496: . bs - size of block
3497: . m - number of rows
3498: . n - number of columns
3499: . i - row indices
3500: . j - column indices
3501: - a - matrix values
3503: Output Parameter:
3504: . mat - the matrix
3506: Level: advanced
3508: Notes:
3509: The i, j, and a arrays are not copied by this routine, the user must free these arrays
3510: once the matrix is destroyed
3512: You cannot set new nonzero locations into this matrix, that will generate an error.
3514: The i and j indices are 0 based
3516: When block size is greater than 1 the matrix values must be stored using the BAIJ storage format (see the BAIJ code to determine this).
3518: The order of the entries in values is the same as the block compressed sparse row storage format; that is, it is
3519: the same as a three dimensional array in Fortran values(bs,bs,nnz) that contains the first column of the first
3520: block, followed by the second column of the first block etc etc. That is, the blocks are contiguous in memory
3521: with column-major ordering within blocks.
3523: .seealso: MatCreate(), MatCreateBAIJ(), MatCreateSeqBAIJ()
3525: @*/
3526: PetscErrorCode MatCreateSeqBAIJWithArrays(MPI_Comm comm,PetscInt bs,PetscInt m,PetscInt n,PetscInt i[],PetscInt j[],PetscScalar a[],Mat *mat)
3527: {
3529: PetscInt ii;
3530: Mat_SeqBAIJ *baij;
3533: if (bs != 1) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_SUP,"block size %D > 1 is not supported yet",bs);
3534: if (m > 0 && i[0]) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"i (row indices) must start with 0");
3536: MatCreate(comm,mat);
3537: MatSetSizes(*mat,m,n,m,n);
3538: MatSetType(*mat,MATSEQBAIJ);
3539: MatSeqBAIJSetPreallocation(*mat,bs,MAT_SKIP_ALLOCATION,0);
3540: baij = (Mat_SeqBAIJ*)(*mat)->data;
3541: PetscMalloc2(m,&baij->imax,m,&baij->ilen);
3542: PetscLogObjectMemory((PetscObject)*mat,2*m*sizeof(PetscInt));
3544: baij->i = i;
3545: baij->j = j;
3546: baij->a = a;
3548: baij->singlemalloc = PETSC_FALSE;
3549: baij->nonew = -1; /*this indicates that inserting a new value in the matrix that generates a new nonzero is an error*/
3550: baij->free_a = PETSC_FALSE;
3551: baij->free_ij = PETSC_FALSE;
3553: for (ii=0; ii<m; ii++) {
3554: baij->ilen[ii] = baij->imax[ii] = i[ii+1] - i[ii];
3555: #if defined(PETSC_USE_DEBUG)
3556: if (i[ii+1] - i[ii] < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative row length in i (row indices) row = %d length = %d",ii,i[ii+1] - i[ii]);
3557: #endif
3558: }
3559: #if defined(PETSC_USE_DEBUG)
3560: for (ii=0; ii<baij->i[m]; ii++) {
3561: if (j[ii] < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Negative column index at location = %d index = %d",ii,j[ii]);
3562: if (j[ii] > n - 1) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Column index to large at location = %d index = %d",ii,j[ii]);
3563: }
3564: #endif
3566: MatAssemblyBegin(*mat,MAT_FINAL_ASSEMBLY);
3567: MatAssemblyEnd(*mat,MAT_FINAL_ASSEMBLY);
3568: return(0);
3569: }
3571: PetscErrorCode MatCreateMPIMatConcatenateSeqMat_SeqBAIJ(MPI_Comm comm,Mat inmat,PetscInt n,MatReuse scall,Mat *outmat)
3572: {
3574: PetscMPIInt size;
3577: MPI_Comm_size(comm,&size);
3578: if (size == 1 && scall == MAT_REUSE_MATRIX) {
3579: MatCopy(inmat,*outmat,SAME_NONZERO_PATTERN);
3580: } else {
3581: MatCreateMPIMatConcatenateSeqMat_MPIBAIJ(comm,inmat,n,scall,outmat);
3582: }
3583: return(0);
3584: }