Actual source code: baij.c
1: #define PETSCMAT_DLL
3: /*
4: Defines the basic matrix operations for the BAIJ (compressed row)
5: matrix storage format.
6: */
7: #include src/mat/impls/baij/seq/baij.h
8: #include src/inline/spops.h
9: #include petscsys.h
11: #include src/inline/ilu.h
15: /*@C
16: MatSeqBAIJInvertBlockDiagonal - Inverts the block diagonal entries.
18: Collective on Mat
20: Input Parameters:
21: . mat - the matrix
23: Level: advanced
24: @*/
25: PetscErrorCode PETSCMAT_DLLEXPORT MatSeqBAIJInvertBlockDiagonal(Mat mat)
26: {
27: PetscErrorCode ierr,(*f)(Mat);
31: if (!mat->assembled) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Not for unassembled matrix");
32: if (mat->factor) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
34: PetscObjectQueryFunction((PetscObject)mat,"MatSeqBAIJInvertBlockDiagonal_C",(void (**)(void))&f);
35: if (f) {
36: (*f)(mat);
37: } else {
38: SETERRQ(PETSC_ERR_SUP,"Currently only implemented for SeqBAIJ.");
39: }
40: return(0);
41: }
46: PetscErrorCode PETSCMAT_DLLEXPORT MatInvertBlockDiagonal_SeqBAIJ(Mat A)
47: {
48: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*) A->data;
50: PetscInt *diag_offset,i,bs = A->bs,mbs = a->mbs;
51: PetscScalar *v = a->a,*odiag,*diag,*mdiag;
54: if (a->idiagvalid) return(0);
55: MatMarkDiagonal_SeqBAIJ(A);
56: diag_offset = a->diag;
57: if (!a->idiag) {
58: PetscMalloc(2*bs*bs*mbs*sizeof(PetscScalar),&a->idiag);
59: }
60: diag = a->idiag;
61: mdiag = a->idiag+bs*bs*mbs;
62: /* factor and invert each block */
63: switch (bs){
64: case 2:
65: for (i=0; i<mbs; i++) {
66: odiag = v + 4*diag_offset[i];
67: diag[0] = odiag[0]; diag[1] = odiag[1]; diag[2] = odiag[2]; diag[3] = odiag[3];
68: mdiag[0] = odiag[0]; mdiag[1] = odiag[1]; mdiag[2] = odiag[2]; mdiag[3] = odiag[3];
69: Kernel_A_gets_inverse_A_2(diag);
70: diag += 4;
71: mdiag += 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: mdiag[0] = odiag[0]; mdiag[1] = odiag[1]; mdiag[2] = odiag[2]; mdiag[3] = odiag[3];
81: mdiag[4] = odiag[4]; mdiag[5] = odiag[5]; mdiag[6] = odiag[6]; mdiag[7] = odiag[7];
82: mdiag[8] = odiag[8];
83: Kernel_A_gets_inverse_A_3(diag);
84: diag += 9;
85: mdiag += 9;
86: }
87: break;
88: case 4:
89: for (i=0; i<mbs; i++) {
90: odiag = v + 16*diag_offset[i];
91: PetscMemcpy(diag,odiag,16*sizeof(PetscScalar));
92: PetscMemcpy(mdiag,odiag,16*sizeof(PetscScalar));
93: Kernel_A_gets_inverse_A_4(diag);
94: diag += 16;
95: mdiag += 16;
96: }
97: break;
98: case 5:
99: for (i=0; i<mbs; i++) {
100: odiag = v + 25*diag_offset[i];
101: PetscMemcpy(diag,odiag,25*sizeof(PetscScalar));
102: PetscMemcpy(mdiag,odiag,25*sizeof(PetscScalar));
103: Kernel_A_gets_inverse_A_5(diag);
104: diag += 25;
105: mdiag += 25;
106: }
107: break;
108: default:
109: SETERRQ1(PETSC_ERR_SUP,"not supported for block size %D",bs);
110: }
111: a->idiagvalid = PETSC_TRUE;
112: return(0);
113: }
118: PetscErrorCode MatPBRelax_SeqBAIJ_2(Mat A,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,PetscInt its,PetscInt lits,Vec xx)
119: {
120: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
121: PetscScalar *x,x1,x2,s1,s2;
122: const PetscScalar *v,*aa = a->a, *b, *idiag,*mdiag;
123: PetscErrorCode ierr;
124: PetscInt m = a->mbs,i,i2,nz,idx;
125: const PetscInt *diag,*ai = a->i,*aj = a->j,*vi;
128: its = its*lits;
129: if (its <= 0) SETERRQ2(PETSC_ERR_ARG_WRONG,"Relaxation requires global its %D and local its %D both positive",its,lits);
130: if (fshift) SETERRQ(PETSC_ERR_SUP,"Sorry, no support for diagonal shift");
131: if (omega != 1.0) SETERRQ(PETSC_ERR_SUP,"Sorry, no support for non-trivial relaxation factor");
132: if ((flag & SOR_EISENSTAT) ||(flag & SOR_APPLY_UPPER) || (flag & SOR_APPLY_LOWER) ) SETERRQ(PETSC_ERR_SUP,"Sorry, no support for Eisenstat trick");
133: if (its > 1) SETERRQ(PETSC_ERR_SUP,"Sorry, no support yet for multiple point block SOR iterations");
135: if (!a->idiagvalid){MatInvertBlockDiagonal_SeqBAIJ(A);}
137: diag = a->diag;
138: idiag = a->idiag;
139: VecGetArray(xx,&x);
140: VecGetArray(bb,(PetscScalar**)&b);
142: if (flag & SOR_ZERO_INITIAL_GUESS) {
143: if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP){
144: x[0] = b[0]*idiag[0] + b[1]*idiag[2];
145: x[1] = b[0]*idiag[1] + b[1]*idiag[3];
146: i2 = 2;
147: idiag += 4;
148: for (i=1; i<m; i++) {
149: v = aa + 4*ai[i];
150: vi = aj + ai[i];
151: nz = diag[i] - ai[i];
152: s1 = b[i2]; s2 = b[i2+1];
153: while (nz--) {
154: idx = 2*(*vi++);
155: x1 = x[idx]; x2 = x[1+idx];
156: s1 -= v[0]*x1 + v[2]*x2;
157: s2 -= v[1]*x1 + v[3]*x2;
158: v += 4;
159: }
160: x[i2] = idiag[0]*s1 + idiag[2]*s2;
161: x[i2+1] = idiag[1]*s1 + idiag[3]*s2;
162: idiag += 4;
163: i2 += 2;
164: }
165: /* for logging purposes assume number of nonzero in lower half is 1/2 of total */
166: PetscLogFlops(4*(a->nz));
167: }
168: if ((flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) &&
169: (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP)) {
170: i2 = 0;
171: mdiag = a->idiag+4*a->mbs;
172: for (i=0; i<m; i++) {
173: x1 = x[i2]; x2 = x[i2+1];
174: x[i2] = mdiag[0]*x1 + mdiag[2]*x2;
175: x[i2+1] = mdiag[1]*x1 + mdiag[3]*x2;
176: mdiag += 4;
177: i2 += 2;
178: }
179: PetscLogFlops(6*m);
180: } else if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) {
181: PetscMemcpy(x,b,A->m*sizeof(PetscScalar));
182: }
183: if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP){
184: idiag = a->idiag+4*a->mbs - 4;
185: i2 = 2*m - 2;
186: x1 = x[i2]; x2 = x[i2+1];
187: x[i2] = idiag[0]*x1 + idiag[2]*x2;
188: x[i2+1] = idiag[1]*x1 + idiag[3]*x2;
189: idiag -= 4;
190: i2 -= 2;
191: for (i=m-2; i>=0; i--) {
192: v = aa + 4*(diag[i]+1);
193: vi = aj + diag[i] + 1;
194: nz = ai[i+1] - diag[i] - 1;
195: s1 = x[i2]; s2 = x[i2+1];
196: while (nz--) {
197: idx = 2*(*vi++);
198: x1 = x[idx]; x2 = x[1+idx];
199: s1 -= v[0]*x1 + v[2]*x2;
200: s2 -= v[1]*x1 + v[3]*x2;
201: v += 4;
202: }
203: x[i2] = idiag[0]*s1 + idiag[2]*s2;
204: x[i2+1] = idiag[1]*s1 + idiag[3]*s2;
205: idiag -= 4;
206: i2 -= 2;
207: }
208: PetscLogFlops(4*(a->nz));
209: }
210: } else {
211: SETERRQ(PETSC_ERR_SUP,"Only supports point block SOR with zero initial guess");
212: }
213: VecRestoreArray(xx,&x);
214: VecRestoreArray(bb,(PetscScalar**)&b);
215: return(0);
216: }
220: PetscErrorCode MatPBRelax_SeqBAIJ_3(Mat A,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,PetscInt its,PetscInt lits,Vec xx)
221: {
222: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
223: PetscScalar *x,x1,x2,x3,s1,s2,s3;
224: const PetscScalar *v,*aa = a->a, *b, *idiag,*mdiag;
225: PetscErrorCode ierr;
226: PetscInt m = a->mbs,i,i2,nz,idx;
227: const PetscInt *diag,*ai = a->i,*aj = a->j,*vi;
230: its = its*lits;
231: if (its <= 0) SETERRQ2(PETSC_ERR_ARG_WRONG,"Relaxation requires global its %D and local its %D both positive",its,lits);
232: if (fshift) SETERRQ(PETSC_ERR_SUP,"Sorry, no support for diagonal shift");
233: if (omega != 1.0) SETERRQ(PETSC_ERR_SUP,"Sorry, no support for non-trivial relaxation factor");
234: if ((flag & SOR_EISENSTAT) ||(flag & SOR_APPLY_UPPER) || (flag & SOR_APPLY_LOWER) ) SETERRQ(PETSC_ERR_SUP,"Sorry, no support for Eisenstat trick");
235: if (its > 1) SETERRQ(PETSC_ERR_SUP,"Sorry, no support yet for multiple point block SOR iterations");
237: if (!a->idiagvalid){MatInvertBlockDiagonal_SeqBAIJ(A);}
239: diag = a->diag;
240: idiag = a->idiag;
241: VecGetArray(xx,&x);
242: VecGetArray(bb,(PetscScalar**)&b);
244: if (flag & SOR_ZERO_INITIAL_GUESS) {
245: if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP){
246: x[0] = b[0]*idiag[0] + b[1]*idiag[3] + b[2]*idiag[6];
247: x[1] = b[0]*idiag[1] + b[1]*idiag[4] + b[2]*idiag[7];
248: x[2] = b[0]*idiag[2] + b[1]*idiag[5] + b[2]*idiag[8];
249: i2 = 3;
250: idiag += 9;
251: for (i=1; i<m; i++) {
252: v = aa + 9*ai[i];
253: vi = aj + ai[i];
254: nz = diag[i] - ai[i];
255: s1 = b[i2]; s2 = b[i2+1]; s3 = b[i2+2];
256: while (nz--) {
257: idx = 3*(*vi++);
258: x1 = x[idx]; x2 = x[1+idx];x3 = x[2+idx];
259: s1 -= v[0]*x1 + v[3]*x2 + v[6]*x3;
260: s2 -= v[1]*x1 + v[4]*x2 + v[7]*x3;
261: s3 -= v[2]*x1 + v[5]*x2 + v[8]*x3;
262: v += 9;
263: }
264: x[i2] = idiag[0]*s1 + idiag[3]*s2 + idiag[6]*s3;
265: x[i2+1] = idiag[1]*s1 + idiag[4]*s2 + idiag[7]*s3;
266: x[i2+2] = idiag[2]*s1 + idiag[5]*s2 + idiag[8]*s3;
267: idiag += 9;
268: i2 += 3;
269: }
270: /* for logging purposes assume number of nonzero in lower half is 1/2 of total */
271: PetscLogFlops(9*(a->nz));
272: }
273: if ((flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) &&
274: (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP)) {
275: i2 = 0;
276: mdiag = a->idiag+9*a->mbs;
277: for (i=0; i<m; i++) {
278: x1 = x[i2]; x2 = x[i2+1]; x3 = x[i2+2];
279: x[i2] = mdiag[0]*x1 + mdiag[3]*x2 + mdiag[6]*x3;
280: x[i2+1] = mdiag[1]*x1 + mdiag[4]*x2 + mdiag[7]*x3;
281: x[i2+2] = mdiag[2]*x1 + mdiag[5]*x2 + mdiag[8]*x3;
282: mdiag += 9;
283: i2 += 3;
284: }
285: PetscLogFlops(15*m);
286: } else if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) {
287: PetscMemcpy(x,b,A->m*sizeof(PetscScalar));
288: }
289: if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP){
290: idiag = a->idiag+9*a->mbs - 9;
291: i2 = 3*m - 3;
292: x1 = x[i2]; x2 = x[i2+1]; x3 = x[i2+2];
293: x[i2] = idiag[0]*x1 + idiag[3]*x2 + idiag[6]*x3;
294: x[i2+1] = idiag[1]*x1 + idiag[4]*x2 + idiag[7]*x3;
295: x[i2+2] = idiag[2]*x1 + idiag[5]*x2 + idiag[8]*x3;
296: idiag -= 9;
297: i2 -= 3;
298: for (i=m-2; i>=0; i--) {
299: v = aa + 9*(diag[i]+1);
300: vi = aj + diag[i] + 1;
301: nz = ai[i+1] - diag[i] - 1;
302: s1 = x[i2]; s2 = x[i2+1]; s3 = x[i2+2];
303: while (nz--) {
304: idx = 3*(*vi++);
305: x1 = x[idx]; x2 = x[1+idx]; x3 = x[2+idx];
306: s1 -= v[0]*x1 + v[3]*x2 + v[6]*x3;
307: s2 -= v[1]*x1 + v[4]*x2 + v[7]*x3;
308: s3 -= v[2]*x1 + v[5]*x2 + v[8]*x3;
309: v += 9;
310: }
311: x[i2] = idiag[0]*s1 + idiag[3]*s2 + idiag[6]*s3;
312: x[i2+1] = idiag[1]*s1 + idiag[4]*s2 + idiag[7]*s3;
313: x[i2+2] = idiag[2]*s1 + idiag[5]*s2 + idiag[8]*s3;
314: idiag -= 9;
315: i2 -= 3;
316: }
317: PetscLogFlops(9*(a->nz));
318: }
319: } else {
320: SETERRQ(PETSC_ERR_SUP,"Only supports point block SOR with zero initial guess");
321: }
322: VecRestoreArray(xx,&x);
323: VecRestoreArray(bb,(PetscScalar**)&b);
324: return(0);
325: }
329: PetscErrorCode MatPBRelax_SeqBAIJ_4(Mat A,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,PetscInt its,PetscInt lits,Vec xx)
330: {
331: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
332: PetscScalar *x,x1,x2,x3,x4,s1,s2,s3,s4;
333: const PetscScalar *v,*aa = a->a, *b, *idiag,*mdiag;
334: PetscErrorCode ierr;
335: PetscInt m = a->mbs,i,i2,nz,idx;
336: const PetscInt *diag,*ai = a->i,*aj = a->j,*vi;
339: its = its*lits;
340: if (its <= 0) SETERRQ2(PETSC_ERR_ARG_WRONG,"Relaxation requires global its %D and local its %D both positive",its,lits);
341: if (fshift) SETERRQ(PETSC_ERR_SUP,"Sorry, no support for diagonal shift");
342: if (omega != 1.0) SETERRQ(PETSC_ERR_SUP,"Sorry, no support for non-trivial relaxation factor");
343: if ((flag & SOR_EISENSTAT) ||(flag & SOR_APPLY_UPPER) || (flag & SOR_APPLY_LOWER) ) SETERRQ(PETSC_ERR_SUP,"Sorry, no support for Eisenstat trick");
344: if (its > 1) SETERRQ(PETSC_ERR_SUP,"Sorry, no support yet for multiple point block SOR iterations");
346: if (!a->idiagvalid){MatInvertBlockDiagonal_SeqBAIJ(A);}
348: diag = a->diag;
349: idiag = a->idiag;
350: VecGetArray(xx,&x);
351: VecGetArray(bb,(PetscScalar**)&b);
353: if (flag & SOR_ZERO_INITIAL_GUESS) {
354: if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP){
355: x[0] = b[0]*idiag[0] + b[1]*idiag[4] + b[2]*idiag[8] + b[3]*idiag[12];
356: x[1] = b[0]*idiag[1] + b[1]*idiag[5] + b[2]*idiag[9] + b[3]*idiag[13];
357: x[2] = b[0]*idiag[2] + b[1]*idiag[6] + b[2]*idiag[10] + b[3]*idiag[14];
358: x[3] = b[0]*idiag[3] + b[1]*idiag[7] + b[2]*idiag[11] + b[3]*idiag[15];
359: i2 = 4;
360: idiag += 16;
361: for (i=1; i<m; i++) {
362: v = aa + 16*ai[i];
363: vi = aj + ai[i];
364: nz = diag[i] - ai[i];
365: s1 = b[i2]; s2 = b[i2+1]; s3 = b[i2+2]; s4 = b[i2+3];
366: while (nz--) {
367: idx = 4*(*vi++);
368: x1 = x[idx]; x2 = x[1+idx]; x3 = x[2+idx]; x4 = x[3+idx];
369: s1 -= v[0]*x1 + v[4]*x2 + v[8]*x3 + v[12]*x4;
370: s2 -= v[1]*x1 + v[5]*x2 + v[9]*x3 + v[13]*x4;
371: s3 -= v[2]*x1 + v[6]*x2 + v[10]*x3 + v[14]*x4;
372: s4 -= v[3]*x1 + v[7]*x2 + v[11]*x3 + v[15]*x4;
373: v += 16;
374: }
375: x[i2] = idiag[0]*s1 + idiag[4]*s2 + idiag[8]*s3 + idiag[12]*s4;
376: x[i2+1] = idiag[1]*s1 + idiag[5]*s2 + idiag[9]*s3 + idiag[13]*s4;
377: x[i2+2] = idiag[2]*s1 + idiag[6]*s2 + idiag[10]*s3 + idiag[14]*s4;
378: x[i2+3] = idiag[3]*s1 + idiag[7]*s2 + idiag[11]*s3 + idiag[15]*s4;
379: idiag += 16;
380: i2 += 4;
381: }
382: /* for logging purposes assume number of nonzero in lower half is 1/2 of total */
383: PetscLogFlops(16*(a->nz));
384: }
385: if ((flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) &&
386: (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP)) {
387: i2 = 0;
388: mdiag = a->idiag+16*a->mbs;
389: for (i=0; i<m; i++) {
390: x1 = x[i2]; x2 = x[i2+1]; x3 = x[i2+2]; x4 = x[i2+3];
391: x[i2] = mdiag[0]*x1 + mdiag[4]*x2 + mdiag[8]*x3 + mdiag[12]*x4;
392: x[i2+1] = mdiag[1]*x1 + mdiag[5]*x2 + mdiag[9]*x3 + mdiag[13]*x4;
393: x[i2+2] = mdiag[2]*x1 + mdiag[6]*x2 + mdiag[10]*x3 + mdiag[14]*x4;
394: x[i2+3] = mdiag[3]*x1 + mdiag[7]*x2 + mdiag[11]*x3 + mdiag[15]*x4;
395: mdiag += 16;
396: i2 += 4;
397: }
398: PetscLogFlops(28*m);
399: } else if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) {
400: PetscMemcpy(x,b,A->m*sizeof(PetscScalar));
401: }
402: if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP){
403: idiag = a->idiag+16*a->mbs - 16;
404: i2 = 4*m - 4;
405: x1 = x[i2]; x2 = x[i2+1]; x3 = x[i2+2]; x4 = x[i2+3];
406: x[i2] = idiag[0]*x1 + idiag[4]*x2 + idiag[8]*x3 + idiag[12]*x4;
407: x[i2+1] = idiag[1]*x1 + idiag[5]*x2 + idiag[9]*x3 + idiag[13]*x4;
408: x[i2+2] = idiag[2]*x1 + idiag[6]*x2 + idiag[10]*x3 + idiag[14]*x4;
409: x[i2+3] = idiag[3]*x1 + idiag[7]*x2 + idiag[11]*x3 + idiag[15]*x4;
410: idiag -= 16;
411: i2 -= 4;
412: for (i=m-2; i>=0; i--) {
413: v = aa + 16*(diag[i]+1);
414: vi = aj + diag[i] + 1;
415: nz = ai[i+1] - diag[i] - 1;
416: s1 = x[i2]; s2 = x[i2+1]; s3 = x[i2+2]; s4 = x[i2+3];
417: while (nz--) {
418: idx = 4*(*vi++);
419: x1 = x[idx]; x2 = x[1+idx]; x3 = x[2+idx]; x4 = x[3+idx];
420: s1 -= v[0]*x1 + v[4]*x2 + v[8]*x3 + v[12]*x4;
421: s2 -= v[1]*x1 + v[5]*x2 + v[9]*x3 + v[13]*x4;
422: s3 -= v[2]*x1 + v[6]*x2 + v[10]*x3 + v[14]*x4;
423: s4 -= v[3]*x1 + v[7]*x2 + v[11]*x3 + v[15]*x4;
424: v += 16;
425: }
426: x[i2] = idiag[0]*s1 + idiag[4]*s2 + idiag[8]*s3 + idiag[12]*s4;
427: x[i2+1] = idiag[1]*s1 + idiag[5]*s2 + idiag[9]*s3 + idiag[13]*s4;
428: x[i2+2] = idiag[2]*s1 + idiag[6]*s2 + idiag[10]*s3 + idiag[14]*s4;
429: x[i2+3] = idiag[3]*s1 + idiag[7]*s2 + idiag[11]*s3 + idiag[15]*s4;
430: idiag -= 16;
431: i2 -= 4;
432: }
433: PetscLogFlops(16*(a->nz));
434: }
435: } else {
436: SETERRQ(PETSC_ERR_SUP,"Only supports point block SOR with zero initial guess");
437: }
438: VecRestoreArray(xx,&x);
439: VecRestoreArray(bb,(PetscScalar**)&b);
440: return(0);
441: }
445: PetscErrorCode MatPBRelax_SeqBAIJ_5(Mat A,Vec bb,PetscReal omega,MatSORType flag,PetscReal fshift,PetscInt its,PetscInt lits,Vec xx)
446: {
447: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
448: PetscScalar *x,x1,x2,x3,x4,x5,s1,s2,s3,s4,s5;
449: const PetscScalar *v,*aa = a->a, *b, *idiag,*mdiag;
450: PetscErrorCode ierr;
451: PetscInt m = a->mbs,i,i2,nz,idx;
452: const PetscInt *diag,*ai = a->i,*aj = a->j,*vi;
455: its = its*lits;
456: if (its <= 0) SETERRQ2(PETSC_ERR_ARG_WRONG,"Relaxation requires global its %D and local its %D both positive",its,lits);
457: if (fshift) SETERRQ(PETSC_ERR_SUP,"Sorry, no support for diagonal shift");
458: if (omega != 1.0) SETERRQ(PETSC_ERR_SUP,"Sorry, no support for non-trivial relaxation factor");
459: if ((flag & SOR_EISENSTAT) ||(flag & SOR_APPLY_UPPER) || (flag & SOR_APPLY_LOWER) ) SETERRQ(PETSC_ERR_SUP,"Sorry, no support for Eisenstat trick");
460: if (its > 1) SETERRQ(PETSC_ERR_SUP,"Sorry, no support yet for multiple point block SOR iterations");
462: if (!a->idiagvalid){MatInvertBlockDiagonal_SeqBAIJ(A);}
464: diag = a->diag;
465: idiag = a->idiag;
466: VecGetArray(xx,&x);
467: VecGetArray(bb,(PetscScalar**)&b);
469: if (flag & SOR_ZERO_INITIAL_GUESS) {
470: if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP){
471: x[0] = b[0]*idiag[0] + b[1]*idiag[5] + b[2]*idiag[10] + b[3]*idiag[15] + b[4]*idiag[20];
472: x[1] = b[0]*idiag[1] + b[1]*idiag[6] + b[2]*idiag[11] + b[3]*idiag[16] + b[4]*idiag[21];
473: x[2] = b[0]*idiag[2] + b[1]*idiag[7] + b[2]*idiag[12] + b[3]*idiag[17] + b[4]*idiag[22];
474: x[3] = b[0]*idiag[3] + b[1]*idiag[8] + b[2]*idiag[13] + b[3]*idiag[18] + b[4]*idiag[23];
475: x[4] = b[0]*idiag[4] + b[1]*idiag[9] + b[2]*idiag[14] + b[3]*idiag[19] + b[4]*idiag[24];
476: i2 = 5;
477: idiag += 25;
478: for (i=1; i<m; i++) {
479: v = aa + 25*ai[i];
480: vi = aj + ai[i];
481: nz = diag[i] - ai[i];
482: s1 = b[i2]; s2 = b[i2+1]; s3 = b[i2+2]; s4 = b[i2+3]; s5 = b[i2+4];
483: while (nz--) {
484: idx = 5*(*vi++);
485: x1 = x[idx]; x2 = x[1+idx]; x3 = x[2+idx]; x4 = x[3+idx]; x5 = x[4+idx];
486: s1 -= v[0]*x1 + v[5]*x2 + v[10]*x3 + v[15]*x4 + v[20]*x5;
487: s2 -= v[1]*x1 + v[6]*x2 + v[11]*x3 + v[16]*x4 + v[21]*x5;
488: s3 -= v[2]*x1 + v[7]*x2 + v[12]*x3 + v[17]*x4 + v[22]*x5;
489: s4 -= v[3]*x1 + v[8]*x2 + v[13]*x3 + v[18]*x4 + v[23]*x5;
490: s5 -= v[4]*x1 + v[9]*x2 + v[14]*x3 + v[19]*x4 + v[24]*x5;
491: v += 25;
492: }
493: x[i2] = idiag[0]*s1 + idiag[5]*s2 + idiag[10]*s3 + idiag[15]*s4 + idiag[20]*s5;
494: x[i2+1] = idiag[1]*s1 + idiag[6]*s2 + idiag[11]*s3 + idiag[16]*s4 + idiag[21]*s5;
495: x[i2+2] = idiag[2]*s1 + idiag[7]*s2 + idiag[12]*s3 + idiag[17]*s4 + idiag[22]*s5;
496: x[i2+3] = idiag[3]*s1 + idiag[8]*s2 + idiag[13]*s3 + idiag[18]*s4 + idiag[23]*s5;
497: x[i2+4] = idiag[4]*s1 + idiag[9]*s2 + idiag[14]*s3 + idiag[19]*s4 + idiag[24]*s5;
498: idiag += 25;
499: i2 += 5;
500: }
501: /* for logging purposes assume number of nonzero in lower half is 1/2 of total */
502: PetscLogFlops(25*(a->nz));
503: }
504: if ((flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) &&
505: (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP)) {
506: i2 = 0;
507: mdiag = a->idiag+25*a->mbs;
508: for (i=0; i<m; i++) {
509: x1 = x[i2]; x2 = x[i2+1]; x3 = x[i2+2]; x4 = x[i2+3]; x5 = x[i2+4];
510: x[i2] = mdiag[0]*x1 + mdiag[5]*x2 + mdiag[10]*x3 + mdiag[15]*x4 + mdiag[20]*x5;
511: x[i2+1] = mdiag[1]*x1 + mdiag[6]*x2 + mdiag[11]*x3 + mdiag[16]*x4 + mdiag[21]*x5;
512: x[i2+2] = mdiag[2]*x1 + mdiag[7]*x2 + mdiag[12]*x3 + mdiag[17]*x4 + mdiag[22]*x5;
513: x[i2+3] = mdiag[3]*x1 + mdiag[8]*x2 + mdiag[13]*x3 + mdiag[18]*x4 + mdiag[23]*x5;
514: x[i2+4] = mdiag[4]*x1 + mdiag[9]*x2 + mdiag[14]*x3 + mdiag[19]*x4 + mdiag[24]*x5;
515: mdiag += 25;
516: i2 += 5;
517: }
518: PetscLogFlops(45*m);
519: } else if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) {
520: PetscMemcpy(x,b,A->m*sizeof(PetscScalar));
521: }
522: if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP){
523: idiag = a->idiag+25*a->mbs - 25;
524: i2 = 5*m - 5;
525: x1 = x[i2]; x2 = x[i2+1]; x3 = x[i2+2]; x4 = x[i2+3]; x5 = x[i2+4];
526: x[i2] = idiag[0]*x1 + idiag[5]*x2 + idiag[10]*x3 + idiag[15]*x4 + idiag[20]*x5;
527: x[i2+1] = idiag[1]*x1 + idiag[6]*x2 + idiag[11]*x3 + idiag[16]*x4 + idiag[21]*x5;
528: x[i2+2] = idiag[2]*x1 + idiag[7]*x2 + idiag[12]*x3 + idiag[17]*x4 + idiag[22]*x5;
529: x[i2+3] = idiag[3]*x1 + idiag[8]*x2 + idiag[13]*x3 + idiag[18]*x4 + idiag[23]*x5;
530: x[i2+4] = idiag[4]*x1 + idiag[9]*x2 + idiag[14]*x3 + idiag[19]*x4 + idiag[24]*x5;
531: idiag -= 25;
532: i2 -= 5;
533: for (i=m-2; i>=0; i--) {
534: v = aa + 25*(diag[i]+1);
535: vi = aj + diag[i] + 1;
536: nz = ai[i+1] - diag[i] - 1;
537: s1 = x[i2]; s2 = x[i2+1]; s3 = x[i2+2]; s4 = x[i2+3]; s5 = x[i2+4];
538: while (nz--) {
539: idx = 5*(*vi++);
540: x1 = x[idx]; x2 = x[1+idx]; x3 = x[2+idx]; x4 = x[3+idx]; x5 = x[4+idx];
541: s1 -= v[0]*x1 + v[5]*x2 + v[10]*x3 + v[15]*x4 + v[20]*x5;
542: s2 -= v[1]*x1 + v[6]*x2 + v[11]*x3 + v[16]*x4 + v[21]*x5;
543: s3 -= v[2]*x1 + v[7]*x2 + v[12]*x3 + v[17]*x4 + v[22]*x5;
544: s4 -= v[3]*x1 + v[8]*x2 + v[13]*x3 + v[18]*x4 + v[23]*x5;
545: s5 -= v[4]*x1 + v[9]*x2 + v[14]*x3 + v[19]*x4 + v[24]*x5;
546: v += 25;
547: }
548: x[i2] = idiag[0]*s1 + idiag[5]*s2 + idiag[10]*s3 + idiag[15]*s4 + idiag[20]*s5;
549: x[i2+1] = idiag[1]*s1 + idiag[6]*s2 + idiag[11]*s3 + idiag[16]*s4 + idiag[21]*s5;
550: x[i2+2] = idiag[2]*s1 + idiag[7]*s2 + idiag[12]*s3 + idiag[17]*s4 + idiag[22]*s5;
551: x[i2+3] = idiag[3]*s1 + idiag[8]*s2 + idiag[13]*s3 + idiag[18]*s4 + idiag[23]*s5;
552: x[i2+4] = idiag[4]*s1 + idiag[9]*s2 + idiag[14]*s3 + idiag[19]*s4 + idiag[24]*s5;
553: idiag -= 25;
554: i2 -= 5;
555: }
556: PetscLogFlops(25*(a->nz));
557: }
558: } else {
559: SETERRQ(PETSC_ERR_SUP,"Only supports point block SOR with zero initial guess");
560: }
561: VecRestoreArray(xx,&x);
562: VecRestoreArray(bb,(PetscScalar**)&b);
563: return(0);
564: }
566: /*
567: Special version for Fun3d sequential benchmark
568: */
569: #if defined(PETSC_HAVE_FORTRAN_CAPS)
570: #define matsetvaluesblocked4_ MATSETVALUESBLOCKED4
571: #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE)
572: #define matsetvaluesblocked4_ matsetvaluesblocked4
573: #endif
578: void PETSCMAT_DLLEXPORT matsetvaluesblocked4_(Mat *AA,PetscInt *mm,const PetscInt im[],PetscInt *nn,const PetscInt in[],const PetscScalar v[])
579: {
580: Mat A = *AA;
581: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
582: PetscInt *rp,k,low,high,t,ii,jj,row,nrow,i,col,l,N,m = *mm,n = *nn;
583: PetscInt *ai=a->i,*ailen=a->ilen;
584: PetscInt *aj=a->j,stepval;
585: const PetscScalar *value = v;
586: MatScalar *ap,*aa = a->a,*bap;
589: stepval = (n-1)*4;
590: for (k=0; k<m; k++) { /* loop over added rows */
591: row = im[k];
592: rp = aj + ai[row];
593: ap = aa + 16*ai[row];
594: nrow = ailen[row];
595: low = 0;
596: for (l=0; l<n; l++) { /* loop over added columns */
597: col = in[l];
598: value = v + k*(stepval+4)*4 + l*4;
599: low = 0; high = nrow;
600: while (high-low > 7) {
601: t = (low+high)/2;
602: if (rp[t] > col) high = t;
603: else low = t;
604: }
605: for (i=low; i<high; i++) {
606: if (rp[i] > col) break;
607: if (rp[i] == col) {
608: bap = ap + 16*i;
609: for (ii=0; ii<4; ii++,value+=stepval) {
610: for (jj=ii; jj<16; jj+=4) {
611: bap[jj] += *value++;
612: }
613: }
614: goto noinsert2;
615: }
616: }
617: N = nrow++ - 1;
618: /* shift up all the later entries in this row */
619: for (ii=N; ii>=i; ii--) {
620: rp[ii+1] = rp[ii];
621: PetscMemcpy(ap+16*(ii+1),ap+16*(ii),16*sizeof(MatScalar));
622: }
623: if (N >= i) {
624: PetscMemzero(ap+16*i,16*sizeof(MatScalar));
625: }
626: rp[i] = col;
627: bap = ap + 16*i;
628: for (ii=0; ii<4; ii++,value+=stepval) {
629: for (jj=ii; jj<16; jj+=4) {
630: bap[jj] = *value++;
631: }
632: }
633: noinsert2:;
634: low = i;
635: }
636: ailen[row] = nrow;
637: }
638: PetscFunctionReturnVoid();
639: }
642: #if defined(PETSC_HAVE_FORTRAN_CAPS)
643: #define matsetvalues4_ MATSETVALUES4
644: #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE)
645: #define matsetvalues4_ matsetvalues4
646: #endif
651: void PETSCMAT_DLLEXPORT matsetvalues4_(Mat *AA,PetscInt *mm,PetscInt *im,PetscInt *nn,PetscInt *in,PetscScalar *v)
652: {
653: Mat A = *AA;
654: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
655: PetscInt *rp,k,low,high,t,ii,row,nrow,i,col,l,N,n = *nn,m = *mm;
656: PetscInt *ai=a->i,*ailen=a->ilen;
657: PetscInt *aj=a->j,brow,bcol;
658: PetscInt ridx,cidx;
659: MatScalar *ap,value,*aa=a->a,*bap;
660:
662: for (k=0; k<m; k++) { /* loop over added rows */
663: row = im[k]; brow = row/4;
664: rp = aj + ai[brow];
665: ap = aa + 16*ai[brow];
666: nrow = ailen[brow];
667: low = 0;
668: for (l=0; l<n; l++) { /* loop over added columns */
669: col = in[l]; bcol = col/4;
670: ridx = row % 4; cidx = col % 4;
671: value = v[l + k*n];
672: low = 0; high = nrow;
673: while (high-low > 7) {
674: t = (low+high)/2;
675: if (rp[t] > bcol) high = t;
676: else low = t;
677: }
678: for (i=low; i<high; i++) {
679: if (rp[i] > bcol) break;
680: if (rp[i] == bcol) {
681: bap = ap + 16*i + 4*cidx + ridx;
682: *bap += value;
683: goto noinsert1;
684: }
685: }
686: N = nrow++ - 1;
687: /* shift up all the later entries in this row */
688: for (ii=N; ii>=i; ii--) {
689: rp[ii+1] = rp[ii];
690: PetscMemcpy(ap+16*(ii+1),ap+16*(ii),16*sizeof(MatScalar));
691: }
692: if (N>=i) {
693: PetscMemzero(ap+16*i,16*sizeof(MatScalar));
694: }
695: rp[i] = bcol;
696: ap[16*i + 4*cidx + ridx] = value;
697: noinsert1:;
698: low = i;
699: }
700: ailen[brow] = nrow;
701: }
702: PetscFunctionReturnVoid();
703: }
706: /* UGLY, ugly, ugly
707: When MatScalar == PetscScalar the function MatSetValuesBlocked_SeqBAIJ_MatScalar() does
708: not exist. Otherwise ..._MatScalar() takes matrix dlements in single precision and
709: inserts them into the single precision data structure. The function MatSetValuesBlocked_SeqBAIJ()
710: converts the entries into single precision and then calls ..._MatScalar() to put them
711: into the single precision data structures.
712: */
713: #if defined(PETSC_USE_MAT_SINGLE)
714: EXTERN PetscErrorCode MatSetValuesBlocked_SeqBAIJ_MatScalar(Mat,PetscInt,const PetscInt[],PetscInt,const PetscInt[],const MatScalar[],InsertMode);
715: #else
716: #define MatSetValuesBlocked_SeqBAIJ_MatScalar MatSetValuesBlocked_SeqBAIJ
717: #endif
719: #define CHUNKSIZE 10
721: /*
722: Checks for missing diagonals
723: */
726: PetscErrorCode MatMissingDiagonal_SeqBAIJ(Mat A)
727: {
728: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
730: PetscInt *diag,*jj = a->j,i;
733: MatMarkDiagonal_SeqBAIJ(A);
734: diag = a->diag;
735: for (i=0; i<a->mbs; i++) {
736: if (jj[diag[i]] != i) {
737: SETERRQ1(PETSC_ERR_PLIB,"Matrix is missing diagonal number %D",i);
738: }
739: }
740: return(0);
741: }
745: PetscErrorCode MatMarkDiagonal_SeqBAIJ(Mat A)
746: {
747: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
749: PetscInt i,j,*diag,m = a->mbs;
752: if (a->diag) return(0);
754: PetscMalloc((m+1)*sizeof(PetscInt),&diag);
755: PetscLogObjectMemory(A,(m+1)*sizeof(PetscInt));
756: for (i=0; i<m; i++) {
757: diag[i] = a->i[i+1];
758: for (j=a->i[i]; j<a->i[i+1]; j++) {
759: if (a->j[j] == i) {
760: diag[i] = j;
761: break;
762: }
763: }
764: }
765: a->diag = diag;
766: return(0);
767: }
770: EXTERN PetscErrorCode MatToSymmetricIJ_SeqAIJ(PetscInt,PetscInt*,PetscInt*,PetscInt,PetscInt,PetscInt**,PetscInt**);
774: static PetscErrorCode MatGetRowIJ_SeqBAIJ(Mat A,PetscInt oshift,PetscTruth symmetric,PetscInt *nn,PetscInt *ia[],PetscInt *ja[],PetscTruth *done)
775: {
776: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
778: PetscInt n = a->mbs,i;
781: *nn = n;
782: if (!ia) return(0);
783: if (symmetric) {
784: MatToSymmetricIJ_SeqAIJ(n,a->i,a->j,0,oshift,ia,ja);
785: } else if (oshift == 1) {
786: /* temporarily add 1 to i and j indices */
787: PetscInt nz = a->i[n];
788: for (i=0; i<nz; i++) a->j[i]++;
789: for (i=0; i<n+1; i++) a->i[i]++;
790: *ia = a->i; *ja = a->j;
791: } else {
792: *ia = a->i; *ja = a->j;
793: }
795: return(0);
796: }
800: static PetscErrorCode MatRestoreRowIJ_SeqBAIJ(Mat A,PetscInt oshift,PetscTruth symmetric,PetscInt *nn,PetscInt *ia[],PetscInt *ja[],PetscTruth *done)
801: {
802: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
804: PetscInt i,n = a->mbs;
807: if (!ia) return(0);
808: if (symmetric) {
809: PetscFree(*ia);
810: PetscFree(*ja);
811: } else if (oshift == 1) {
812: PetscInt nz = a->i[n]-1;
813: for (i=0; i<nz; i++) a->j[i]--;
814: for (i=0; i<n+1; i++) a->i[i]--;
815: }
816: return(0);
817: }
821: PetscErrorCode MatDestroy_SeqBAIJ(Mat A)
822: {
823: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
827: #if defined(PETSC_USE_LOG)
828: PetscLogObjectState((PetscObject)A,"Rows=%D, Cols=%D, NZ=%D",A->m,A->n,a->nz);
829: #endif
830: MatSeqXAIJFreeAIJ(a->singlemalloc,&a->a,&a->j,&a->i);
831: if (a->row) {
832: ISDestroy(a->row);
833: }
834: if (a->col) {
835: ISDestroy(a->col);
836: }
837: if (a->diag) {PetscFree(a->diag);}
838: if (a->idiag) {PetscFree(a->idiag);}
839: if (a->imax) {PetscFree2(a->imax,a->ilen);}
840: if (a->solve_work) {PetscFree(a->solve_work);}
841: if (a->mult_work) {PetscFree(a->mult_work);}
842: if (a->icol) {ISDestroy(a->icol);}
843: if (a->saved_values) {PetscFree(a->saved_values);}
844: #if defined(PETSC_USE_MAT_SINGLE)
845: if (a->setvaluescopy) {PetscFree(a->setvaluescopy);}
846: #endif
847: if (a->xtoy) {PetscFree(a->xtoy);}
848: if (a->compressedrow.use){PetscFree(a->compressedrow.i);}
850: PetscFree(a);
852: PetscObjectComposeFunction((PetscObject)A,"MatSeqBAIJInvertBlockDiagonal_C","",PETSC_NULL);
853: PetscObjectComposeFunction((PetscObject)A,"MatStoreValues_C","",PETSC_NULL);
854: PetscObjectComposeFunction((PetscObject)A,"MatRetrieveValues_C","",PETSC_NULL);
855: PetscObjectComposeFunction((PetscObject)A,"MatSeqBAIJSetColumnIndices_C","",PETSC_NULL);
856: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqbaij_seqaij_C","",PETSC_NULL);
857: PetscObjectComposeFunction((PetscObject)A,"MatConvert_seqbaij_seqsbaij_C","",PETSC_NULL);
858: PetscObjectComposeFunction((PetscObject)A,"MatSeqBAIJSetPreallocation_C","",PETSC_NULL);
859: return(0);
860: }
864: PetscErrorCode MatSetOption_SeqBAIJ(Mat A,MatOption op)
865: {
866: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
870: switch (op) {
871: case MAT_ROW_ORIENTED:
872: a->roworiented = PETSC_TRUE;
873: break;
874: case MAT_COLUMN_ORIENTED:
875: a->roworiented = PETSC_FALSE;
876: break;
877: case MAT_COLUMNS_SORTED:
878: a->sorted = PETSC_TRUE;
879: break;
880: case MAT_COLUMNS_UNSORTED:
881: a->sorted = PETSC_FALSE;
882: break;
883: case MAT_KEEP_ZEROED_ROWS:
884: a->keepzeroedrows = PETSC_TRUE;
885: break;
886: case MAT_NO_NEW_NONZERO_LOCATIONS:
887: a->nonew = 1;
888: break;
889: case MAT_NEW_NONZERO_LOCATION_ERR:
890: a->nonew = -1;
891: break;
892: case MAT_NEW_NONZERO_ALLOCATION_ERR:
893: a->nonew = -2;
894: break;
895: case MAT_YES_NEW_NONZERO_LOCATIONS:
896: a->nonew = 0;
897: break;
898: case MAT_ROWS_SORTED:
899: case MAT_ROWS_UNSORTED:
900: case MAT_YES_NEW_DIAGONALS:
901: case MAT_IGNORE_OFF_PROC_ENTRIES:
902: case MAT_USE_HASH_TABLE:
903: PetscLogInfo((A,"MatSetOption_SeqBAIJ:Option ignored\n"));
904: break;
905: case MAT_NO_NEW_DIAGONALS:
906: SETERRQ(PETSC_ERR_SUP,"MAT_NO_NEW_DIAGONALS");
907: case MAT_SYMMETRIC:
908: case MAT_STRUCTURALLY_SYMMETRIC:
909: case MAT_NOT_SYMMETRIC:
910: case MAT_NOT_STRUCTURALLY_SYMMETRIC:
911: case MAT_HERMITIAN:
912: case MAT_NOT_HERMITIAN:
913: case MAT_SYMMETRY_ETERNAL:
914: case MAT_NOT_SYMMETRY_ETERNAL:
915: break;
916: default:
917: SETERRQ(PETSC_ERR_SUP,"unknown option");
918: }
919: return(0);
920: }
924: PetscErrorCode MatGetRow_SeqBAIJ(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v)
925: {
926: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
928: PetscInt itmp,i,j,k,M,*ai,*aj,bs,bn,bp,*idx_i,bs2;
929: MatScalar *aa,*aa_i;
930: PetscScalar *v_i;
933: bs = A->bs;
934: ai = a->i;
935: aj = a->j;
936: aa = a->a;
937: bs2 = a->bs2;
938:
939: if (row < 0 || row >= A->m) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"Row %D out of range", row);
940:
941: bn = row/bs; /* Block number */
942: bp = row % bs; /* Block Position */
943: M = ai[bn+1] - ai[bn];
944: *nz = bs*M;
945:
946: if (v) {
947: *v = 0;
948: if (*nz) {
949: PetscMalloc((*nz)*sizeof(PetscScalar),v);
950: for (i=0; i<M; i++) { /* for each block in the block row */
951: v_i = *v + i*bs;
952: aa_i = aa + bs2*(ai[bn] + i);
953: for (j=bp,k=0; j<bs2; j+=bs,k++) {v_i[k] = aa_i[j];}
954: }
955: }
956: }
958: if (idx) {
959: *idx = 0;
960: if (*nz) {
961: PetscMalloc((*nz)*sizeof(PetscInt),idx);
962: for (i=0; i<M; i++) { /* for each block in the block row */
963: idx_i = *idx + i*bs;
964: itmp = bs*aj[ai[bn] + i];
965: for (j=0; j<bs; j++) {idx_i[j] = itmp++;}
966: }
967: }
968: }
969: return(0);
970: }
974: PetscErrorCode MatRestoreRow_SeqBAIJ(Mat A,PetscInt row,PetscInt *nz,PetscInt **idx,PetscScalar **v)
975: {
979: if (idx) {if (*idx) {PetscFree(*idx);}}
980: if (v) {if (*v) {PetscFree(*v);}}
981: return(0);
982: }
986: PetscErrorCode MatTranspose_SeqBAIJ(Mat A,Mat *B)
987: {
988: Mat_SeqBAIJ *a=(Mat_SeqBAIJ *)A->data;
989: Mat C;
991: PetscInt i,j,k,*aj=a->j,*ai=a->i,bs=A->bs,mbs=a->mbs,nbs=a->nbs,len,*col;
992: PetscInt *rows,*cols,bs2=a->bs2;
993: PetscScalar *array;
996: if (!B && mbs!=nbs) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Square matrix only for in-place");
997: PetscMalloc((1+nbs)*sizeof(PetscInt),&col);
998: PetscMemzero(col,(1+nbs)*sizeof(PetscInt));
1000: #if defined(PETSC_USE_MAT_SINGLE)
1001: PetscMalloc(a->bs2*a->nz*sizeof(PetscScalar),&array);
1002: for (i=0; i<a->bs2*a->nz; i++) array[i] = (PetscScalar)a->a[i];
1003: #else
1004: array = a->a;
1005: #endif
1007: for (i=0; i<ai[mbs]; i++) col[aj[i]] += 1;
1008: MatCreate(A->comm,&C);
1009: MatSetSizes(C,A->n,A->m,A->n,A->m);
1010: MatSetType(C,A->type_name);
1011: MatSeqBAIJSetPreallocation_SeqBAIJ(C,bs,PETSC_NULL,col);
1012: PetscFree(col);
1013: PetscMalloc(2*bs*sizeof(PetscInt),&rows);
1014: cols = rows + bs;
1015: for (i=0; i<mbs; i++) {
1016: cols[0] = i*bs;
1017: for (k=1; k<bs; k++) cols[k] = cols[k-1] + 1;
1018: len = ai[i+1] - ai[i];
1019: for (j=0; j<len; j++) {
1020: rows[0] = (*aj++)*bs;
1021: for (k=1; k<bs; k++) rows[k] = rows[k-1] + 1;
1022: MatSetValues(C,bs,rows,bs,cols,array,INSERT_VALUES);
1023: array += bs2;
1024: }
1025: }
1026: PetscFree(rows);
1027: #if defined(PETSC_USE_MAT_SINGLE)
1028: PetscFree(array);
1029: #endif
1030:
1031: MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);
1032: MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);
1033:
1034: if (B) {
1035: *B = C;
1036: } else {
1037: MatHeaderCopy(A,C);
1038: }
1039: return(0);
1040: }
1044: static PetscErrorCode MatView_SeqBAIJ_Binary(Mat A,PetscViewer viewer)
1045: {
1046: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
1048: PetscInt i,*col_lens,bs = A->bs,count,*jj,j,k,l,bs2=a->bs2;
1049: int fd;
1050: PetscScalar *aa;
1051: FILE *file;
1054: PetscViewerBinaryGetDescriptor(viewer,&fd);
1055: PetscMalloc((4+A->m)*sizeof(PetscInt),&col_lens);
1056: col_lens[0] = MAT_FILE_COOKIE;
1058: col_lens[1] = A->m;
1059: col_lens[2] = A->n;
1060: col_lens[3] = a->nz*bs2;
1062: /* store lengths of each row and write (including header) to file */
1063: count = 0;
1064: for (i=0; i<a->mbs; i++) {
1065: for (j=0; j<bs; j++) {
1066: col_lens[4+count++] = bs*(a->i[i+1] - a->i[i]);
1067: }
1068: }
1069: PetscBinaryWrite(fd,col_lens,4+A->m,PETSC_INT,PETSC_TRUE);
1070: PetscFree(col_lens);
1072: /* store column indices (zero start index) */
1073: PetscMalloc((a->nz+1)*bs2*sizeof(PetscInt),&jj);
1074: count = 0;
1075: for (i=0; i<a->mbs; i++) {
1076: for (j=0; j<bs; j++) {
1077: for (k=a->i[i]; k<a->i[i+1]; k++) {
1078: for (l=0; l<bs; l++) {
1079: jj[count++] = bs*a->j[k] + l;
1080: }
1081: }
1082: }
1083: }
1084: PetscBinaryWrite(fd,jj,bs2*a->nz,PETSC_INT,PETSC_FALSE);
1085: PetscFree(jj);
1087: /* store nonzero values */
1088: PetscMalloc((a->nz+1)*bs2*sizeof(PetscScalar),&aa);
1089: count = 0;
1090: for (i=0; i<a->mbs; i++) {
1091: for (j=0; j<bs; j++) {
1092: for (k=a->i[i]; k<a->i[i+1]; k++) {
1093: for (l=0; l<bs; l++) {
1094: aa[count++] = a->a[bs2*k + l*bs + j];
1095: }
1096: }
1097: }
1098: }
1099: PetscBinaryWrite(fd,aa,bs2*a->nz,PETSC_SCALAR,PETSC_FALSE);
1100: PetscFree(aa);
1102: PetscViewerBinaryGetInfoPointer(viewer,&file);
1103: if (file) {
1104: fprintf(file,"-matload_block_size %d\n",(int)A->bs);
1105: }
1106: return(0);
1107: }
1111: static PetscErrorCode MatView_SeqBAIJ_ASCII(Mat A,PetscViewer viewer)
1112: {
1113: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
1114: PetscErrorCode ierr;
1115: PetscInt i,j,bs = A->bs,k,l,bs2=a->bs2;
1116: PetscViewerFormat format;
1119: PetscViewerGetFormat(viewer,&format);
1120: if (format == PETSC_VIEWER_ASCII_INFO || format == PETSC_VIEWER_ASCII_INFO_DETAIL) {
1121: PetscViewerASCIIPrintf(viewer," block size is %D\n",bs);
1122: } else if (format == PETSC_VIEWER_ASCII_MATLAB) {
1123: Mat aij;
1124: MatConvert(A,MATSEQAIJ,MAT_INITIAL_MATRIX,&aij);
1125: MatView(aij,viewer);
1126: MatDestroy(aij);
1127: } else if (format == PETSC_VIEWER_ASCII_FACTOR_INFO) {
1128: return(0);
1129: } else if (format == PETSC_VIEWER_ASCII_COMMON) {
1130: PetscViewerASCIIUseTabs(viewer,PETSC_NO);
1131: for (i=0; i<a->mbs; i++) {
1132: for (j=0; j<bs; j++) {
1133: PetscViewerASCIIPrintf(viewer,"row %D:",i*bs+j);
1134: for (k=a->i[i]; k<a->i[i+1]; k++) {
1135: for (l=0; l<bs; l++) {
1136: #if defined(PETSC_USE_COMPLEX)
1137: if (PetscImaginaryPart(a->a[bs2*k + l*bs + j]) > 0.0 && PetscRealPart(a->a[bs2*k + l*bs + j]) != 0.0) {
1138: PetscViewerASCIIPrintf(viewer," (%D, %g + %gi) ",bs*a->j[k]+l,
1139: PetscRealPart(a->a[bs2*k + l*bs + j]),PetscImaginaryPart(a->a[bs2*k + l*bs + j]));
1140: } else if (PetscImaginaryPart(a->a[bs2*k + l*bs + j]) < 0.0 && PetscRealPart(a->a[bs2*k + l*bs + j]) != 0.0) {
1141: PetscViewerASCIIPrintf(viewer," (%D, %g - %gi) ",bs*a->j[k]+l,
1142: PetscRealPart(a->a[bs2*k + l*bs + j]),-PetscImaginaryPart(a->a[bs2*k + l*bs + j]));
1143: } else if (PetscRealPart(a->a[bs2*k + l*bs + j]) != 0.0) {
1144: PetscViewerASCIIPrintf(viewer," (%D, %g) ",bs*a->j[k]+l,PetscRealPart(a->a[bs2*k + l*bs + j]));
1145: }
1146: #else
1147: if (a->a[bs2*k + l*bs + j] != 0.0) {
1148: PetscViewerASCIIPrintf(viewer," (%D, %g) ",bs*a->j[k]+l,a->a[bs2*k + l*bs + j]);
1149: }
1150: #endif
1151: }
1152: }
1153: PetscViewerASCIIPrintf(viewer,"\n");
1154: }
1155: }
1156: PetscViewerASCIIUseTabs(viewer,PETSC_YES);
1157: } else {
1158: PetscViewerASCIIUseTabs(viewer,PETSC_NO);
1159: for (i=0; i<a->mbs; i++) {
1160: for (j=0; j<bs; j++) {
1161: PetscViewerASCIIPrintf(viewer,"row %D:",i*bs+j);
1162: for (k=a->i[i]; k<a->i[i+1]; k++) {
1163: for (l=0; l<bs; l++) {
1164: #if defined(PETSC_USE_COMPLEX)
1165: if (PetscImaginaryPart(a->a[bs2*k + l*bs + j]) > 0.0) {
1166: PetscViewerASCIIPrintf(viewer," (%D, %g + %g i) ",bs*a->j[k]+l,
1167: PetscRealPart(a->a[bs2*k + l*bs + j]),PetscImaginaryPart(a->a[bs2*k + l*bs + j]));
1168: } else if (PetscImaginaryPart(a->a[bs2*k + l*bs + j]) < 0.0) {
1169: PetscViewerASCIIPrintf(viewer," (%D, %g - %g i) ",bs*a->j[k]+l,
1170: PetscRealPart(a->a[bs2*k + l*bs + j]),-PetscImaginaryPart(a->a[bs2*k + l*bs + j]));
1171: } else {
1172: PetscViewerASCIIPrintf(viewer," (%D, %g) ",bs*a->j[k]+l,PetscRealPart(a->a[bs2*k + l*bs + j]));
1173: }
1174: #else
1175: PetscViewerASCIIPrintf(viewer," (%D, %g) ",bs*a->j[k]+l,a->a[bs2*k + l*bs + j]);
1176: #endif
1177: }
1178: }
1179: PetscViewerASCIIPrintf(viewer,"\n");
1180: }
1181: }
1182: PetscViewerASCIIUseTabs(viewer,PETSC_YES);
1183: }
1184: PetscViewerFlush(viewer);
1185: return(0);
1186: }
1190: static PetscErrorCode MatView_SeqBAIJ_Draw_Zoom(PetscDraw draw,void *Aa)
1191: {
1192: Mat A = (Mat) Aa;
1193: Mat_SeqBAIJ *a=(Mat_SeqBAIJ*)A->data;
1195: PetscInt row,i,j,k,l,mbs=a->mbs,color,bs=A->bs,bs2=a->bs2;
1196: PetscReal xl,yl,xr,yr,x_l,x_r,y_l,y_r;
1197: MatScalar *aa;
1198: PetscViewer viewer;
1202: /* still need to add support for contour plot of nonzeros; see MatView_SeqAIJ_Draw_Zoom()*/
1203: PetscObjectQuery((PetscObject)A,"Zoomviewer",(PetscObject*)&viewer);
1205: PetscDrawGetCoordinates(draw,&xl,&yl,&xr,&yr);
1207: /* loop over matrix elements drawing boxes */
1208: color = PETSC_DRAW_BLUE;
1209: for (i=0,row=0; i<mbs; i++,row+=bs) {
1210: for (j=a->i[i]; j<a->i[i+1]; j++) {
1211: y_l = A->m - row - 1.0; y_r = y_l + 1.0;
1212: x_l = a->j[j]*bs; x_r = x_l + 1.0;
1213: aa = a->a + j*bs2;
1214: for (k=0; k<bs; k++) {
1215: for (l=0; l<bs; l++) {
1216: if (PetscRealPart(*aa++) >= 0.) continue;
1217: PetscDrawRectangle(draw,x_l+k,y_l-l,x_r+k,y_r-l,color,color,color,color);
1218: }
1219: }
1220: }
1221: }
1222: color = PETSC_DRAW_CYAN;
1223: for (i=0,row=0; i<mbs; i++,row+=bs) {
1224: for (j=a->i[i]; j<a->i[i+1]; j++) {
1225: y_l = A->m - row - 1.0; y_r = y_l + 1.0;
1226: x_l = a->j[j]*bs; x_r = x_l + 1.0;
1227: aa = a->a + j*bs2;
1228: for (k=0; k<bs; k++) {
1229: for (l=0; l<bs; l++) {
1230: if (PetscRealPart(*aa++) != 0.) continue;
1231: PetscDrawRectangle(draw,x_l+k,y_l-l,x_r+k,y_r-l,color,color,color,color);
1232: }
1233: }
1234: }
1235: }
1237: color = PETSC_DRAW_RED;
1238: for (i=0,row=0; i<mbs; i++,row+=bs) {
1239: for (j=a->i[i]; j<a->i[i+1]; j++) {
1240: y_l = A->m - row - 1.0; y_r = y_l + 1.0;
1241: x_l = a->j[j]*bs; x_r = x_l + 1.0;
1242: aa = a->a + j*bs2;
1243: for (k=0; k<bs; k++) {
1244: for (l=0; l<bs; l++) {
1245: if (PetscRealPart(*aa++) <= 0.) continue;
1246: PetscDrawRectangle(draw,x_l+k,y_l-l,x_r+k,y_r-l,color,color,color,color);
1247: }
1248: }
1249: }
1250: }
1251: return(0);
1252: }
1256: static PetscErrorCode MatView_SeqBAIJ_Draw(Mat A,PetscViewer viewer)
1257: {
1259: PetscReal xl,yl,xr,yr,w,h;
1260: PetscDraw draw;
1261: PetscTruth isnull;
1265: PetscViewerDrawGetDraw(viewer,0,&draw);
1266: PetscDrawIsNull(draw,&isnull); if (isnull) return(0);
1268: PetscObjectCompose((PetscObject)A,"Zoomviewer",(PetscObject)viewer);
1269: xr = A->n; yr = A->m; h = yr/10.0; w = xr/10.0;
1270: xr += w; yr += h; xl = -w; yl = -h;
1271: PetscDrawSetCoordinates(draw,xl,yl,xr,yr);
1272: PetscDrawZoom(draw,MatView_SeqBAIJ_Draw_Zoom,A);
1273: PetscObjectCompose((PetscObject)A,"Zoomviewer",PETSC_NULL);
1274: return(0);
1275: }
1279: PetscErrorCode MatView_SeqBAIJ(Mat A,PetscViewer viewer)
1280: {
1282: PetscTruth iascii,isbinary,isdraw;
1285: PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_ASCII,&iascii);
1286: PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_BINARY,&isbinary);
1287: PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_DRAW,&isdraw);
1288: if (iascii){
1289: MatView_SeqBAIJ_ASCII(A,viewer);
1290: } else if (isbinary) {
1291: MatView_SeqBAIJ_Binary(A,viewer);
1292: } else if (isdraw) {
1293: MatView_SeqBAIJ_Draw(A,viewer);
1294: } else {
1295: Mat B;
1296: MatConvert(A,MATSEQAIJ,MAT_INITIAL_MATRIX,&B);
1297: MatView(B,viewer);
1298: MatDestroy(B);
1299: }
1300: return(0);
1301: }
1306: PetscErrorCode MatGetValues_SeqBAIJ(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],PetscScalar v[])
1307: {
1308: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
1309: PetscInt *rp,k,low,high,t,row,nrow,i,col,l,*aj = a->j;
1310: PetscInt *ai = a->i,*ailen = a->ilen;
1311: PetscInt brow,bcol,ridx,cidx,bs=A->bs,bs2=a->bs2;
1312: MatScalar *ap,*aa = a->a,zero = 0.0;
1315: for (k=0; k<m; k++) { /* loop over rows */
1316: row = im[k]; brow = row/bs;
1317: if (row < 0) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Negative row");
1318: if (row >= A->m) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"Row %D too large", row);
1319: rp = aj + ai[brow] ; ap = aa + bs2*ai[brow] ;
1320: nrow = ailen[brow];
1321: for (l=0; l<n; l++) { /* loop over columns */
1322: if (in[l] < 0) SETERRQ(PETSC_ERR_ARG_OUTOFRANGE,"Negative column");
1323: if (in[l] >= A->n) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"Column %D too large", in[l]);
1324: col = in[l] ;
1325: bcol = col/bs;
1326: cidx = col%bs;
1327: ridx = row%bs;
1328: high = nrow;
1329: low = 0; /* assume unsorted */
1330: while (high-low > 5) {
1331: t = (low+high)/2;
1332: if (rp[t] > bcol) high = t;
1333: else low = t;
1334: }
1335: for (i=low; i<high; i++) {
1336: if (rp[i] > bcol) break;
1337: if (rp[i] == bcol) {
1338: *v++ = ap[bs2*i+bs*cidx+ridx];
1339: goto finished;
1340: }
1341: }
1342: *v++ = zero;
1343: finished:;
1344: }
1345: }
1346: return(0);
1347: }
1349: #if defined(PETSC_USE_MAT_SINGLE)
1352: PetscErrorCode MatSetValuesBlocked_SeqBAIJ(Mat mat,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode addv)
1353: {
1354: Mat_SeqBAIJ *b = (Mat_SeqBAIJ*)mat->data;
1356: PetscInt i,N = m*n*b->bs2;
1357: MatScalar *vsingle;
1360: if (N > b->setvalueslen) {
1361: if (b->setvaluescopy) {PetscFree(b->setvaluescopy);}
1362: PetscMalloc(N*sizeof(MatScalar),&b->setvaluescopy);
1363: b->setvalueslen = N;
1364: }
1365: vsingle = b->setvaluescopy;
1366: for (i=0; i<N; i++) {
1367: vsingle[i] = v[i];
1368: }
1369: MatSetValuesBlocked_SeqBAIJ_MatScalar(mat,m,im,n,in,vsingle,addv);
1370: return(0);
1371: }
1372: #endif
1377: PetscErrorCode MatSetValuesBlocked_SeqBAIJ_MatScalar(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const MatScalar v[],InsertMode is)
1378: {
1379: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
1380: PetscInt *rp,k,low,high,t,ii,jj,row,nrow,i,col,l,rmax,N,lastcol = -1;
1381: PetscInt *imax=a->imax,*ai=a->i,*ailen=a->ilen;
1382: PetscErrorCode ierr;
1383: PetscInt *aj=a->j,nonew=a->nonew,bs2=a->bs2,bs=A->bs,stepval;
1384: PetscTruth roworiented=a->roworiented;
1385: const MatScalar *value = v;
1386: MatScalar *ap,*aa = a->a,*bap;
1389: if (roworiented) {
1390: stepval = (n-1)*bs;
1391: } else {
1392: stepval = (m-1)*bs;
1393: }
1394: for (k=0; k<m; k++) { /* loop over added rows */
1395: row = im[k];
1396: if (row < 0) continue;
1397: #if defined(PETSC_USE_DEBUG)
1398: if (row >= a->mbs) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",row,a->mbs-1);
1399: #endif
1400: rp = aj + ai[row];
1401: ap = aa + bs2*ai[row];
1402: rmax = imax[row];
1403: nrow = ailen[row];
1404: low = 0;
1405: high = nrow;
1406: for (l=0; l<n; l++) { /* loop over added columns */
1407: if (in[l] < 0) continue;
1408: #if defined(PETSC_USE_DEBUG)
1409: if (in[l] >= a->nbs) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",in[l],a->nbs-1);
1410: #endif
1411: col = in[l];
1412: if (roworiented) {
1413: value = v + k*(stepval+bs)*bs + l*bs;
1414: } else {
1415: value = v + l*(stepval+bs)*bs + k*bs;
1416: }
1417: if (col <= lastcol) low = 0; else high = nrow;
1418: lastcol = col;
1419: while (high-low > 7) {
1420: t = (low+high)/2;
1421: if (rp[t] > col) high = t;
1422: else low = t;
1423: }
1424: for (i=low; i<high; i++) {
1425: if (rp[i] > col) break;
1426: if (rp[i] == col) {
1427: bap = ap + bs2*i;
1428: if (roworiented) {
1429: if (is == ADD_VALUES) {
1430: for (ii=0; ii<bs; ii++,value+=stepval) {
1431: for (jj=ii; jj<bs2; jj+=bs) {
1432: bap[jj] += *value++;
1433: }
1434: }
1435: } else {
1436: for (ii=0; ii<bs; ii++,value+=stepval) {
1437: for (jj=ii; jj<bs2; jj+=bs) {
1438: bap[jj] = *value++;
1439: }
1440: }
1441: }
1442: } else {
1443: if (is == ADD_VALUES) {
1444: for (ii=0; ii<bs; ii++,value+=stepval) {
1445: for (jj=0; jj<bs; jj++) {
1446: *bap++ += *value++;
1447: }
1448: }
1449: } else {
1450: for (ii=0; ii<bs; ii++,value+=stepval) {
1451: for (jj=0; jj<bs; jj++) {
1452: *bap++ = *value++;
1453: }
1454: }
1455: }
1456: }
1457: goto noinsert2;
1458: }
1459: }
1460: if (nonew == 1) goto noinsert2;
1461: if (nonew == -1) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero (%D, %D) in the matrix", row, col);
1462: MatSeqXAIJReallocateAIJ(a,bs2,nrow,row,col,rmax,aa,ai,aj,a->mbs,rp,ap,imax,nonew);
1463: N = nrow++ - 1;
1464: /* shift up all the later entries in this row */
1465: for (ii=N; ii>=i; ii--) {
1466: rp[ii+1] = rp[ii];
1467: PetscMemcpy(ap+bs2*(ii+1),ap+bs2*(ii),bs2*sizeof(MatScalar));
1468: }
1469: if (N >= i) {
1470: PetscMemzero(ap+bs2*i,bs2*sizeof(MatScalar));
1471: }
1472: rp[i] = col;
1473: bap = ap + bs2*i;
1474: if (roworiented) {
1475: for (ii=0; ii<bs; ii++,value+=stepval) {
1476: for (jj=ii; jj<bs2; jj+=bs) {
1477: bap[jj] = *value++;
1478: }
1479: }
1480: } else {
1481: for (ii=0; ii<bs; ii++,value+=stepval) {
1482: for (jj=0; jj<bs; jj++) {
1483: *bap++ = *value++;
1484: }
1485: }
1486: }
1487: noinsert2:;
1488: low = i;
1489: }
1490: ailen[row] = nrow;
1491: }
1492: return(0);
1493: }
1497: PetscErrorCode MatAssemblyEnd_SeqBAIJ(Mat A,MatAssemblyType mode)
1498: {
1499: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
1500: PetscInt fshift = 0,i,j,*ai = a->i,*aj = a->j,*imax = a->imax;
1501: PetscInt m = A->m,*ip,N,*ailen = a->ilen;
1503: PetscInt mbs = a->mbs,bs2 = a->bs2,rmax = 0;
1504: MatScalar *aa = a->a,*ap;
1505: PetscReal ratio=0.6;
1508: if (mode == MAT_FLUSH_ASSEMBLY) return(0);
1510: if (m) rmax = ailen[0];
1511: for (i=1; i<mbs; i++) {
1512: /* move each row back by the amount of empty slots (fshift) before it*/
1513: fshift += imax[i-1] - ailen[i-1];
1514: rmax = PetscMax(rmax,ailen[i]);
1515: if (fshift) {
1516: ip = aj + ai[i]; ap = aa + bs2*ai[i];
1517: N = ailen[i];
1518: for (j=0; j<N; j++) {
1519: ip[j-fshift] = ip[j];
1520: PetscMemcpy(ap+(j-fshift)*bs2,ap+j*bs2,bs2*sizeof(MatScalar));
1521: }
1522: }
1523: ai[i] = ai[i-1] + ailen[i-1];
1524: }
1525: if (mbs) {
1526: fshift += imax[mbs-1] - ailen[mbs-1];
1527: ai[mbs] = ai[mbs-1] + ailen[mbs-1];
1528: }
1529: /* reset ilen and imax for each row */
1530: for (i=0; i<mbs; i++) {
1531: ailen[i] = imax[i] = ai[i+1] - ai[i];
1532: }
1533: a->nz = ai[mbs];
1535: /* diagonals may have moved, so kill the diagonal pointers */
1536: a->idiagvalid = PETSC_FALSE;
1537: if (fshift && a->diag) {
1538: PetscFree(a->diag);
1539: PetscLogObjectMemory(A,-(mbs+1)*sizeof(PetscInt));
1540: a->diag = 0;
1541: }
1542: PetscLogInfo((A,"MatAssemblyEnd_SeqBAIJ:Matrix size: %D X %D, block size %D; storage space: %D unneeded, %D used\n",m,A->n,A->bs,fshift*bs2,a->nz*bs2));
1543: PetscLogInfo((A,"MatAssemblyEnd_SeqBAIJ:Number of mallocs during MatSetValues is %D\n",a->reallocs));
1544: PetscLogInfo((A,"MatAssemblyEnd_SeqBAIJ:Most nonzeros blocks in any row is %D\n",rmax));
1545: a->reallocs = 0;
1546: A->info.nz_unneeded = (PetscReal)fshift*bs2;
1548: /* check for zero rows. If found a large number of zero rows, use CompressedRow functions */
1549: if (a->compressedrow.use){
1550: Mat_CheckCompressedRow(A,&a->compressedrow,a->i,mbs,ratio);
1551: }
1553: A->same_nonzero = PETSC_TRUE;
1554: return(0);
1555: }
1557: /*
1558: This function returns an array of flags which indicate the locations of contiguous
1559: blocks that should be zeroed. for eg: if bs = 3 and is = [0,1,2,3,5,6,7,8,9]
1560: then the resulting sizes = [3,1,1,3,1] correspondig to sets [(0,1,2),(3),(5),(6,7,8),(9)]
1561: Assume: sizes should be long enough to hold all the values.
1562: */
1565: static PetscErrorCode MatZeroRows_SeqBAIJ_Check_Blocks(PetscInt idx[],PetscInt n,PetscInt bs,PetscInt sizes[], PetscInt *bs_max)
1566: {
1567: PetscInt i,j,k,row;
1568: PetscTruth flg;
1571: for (i=0,j=0; i<n; j++) {
1572: row = idx[i];
1573: if (row%bs!=0) { /* Not the begining of a block */
1574: sizes[j] = 1;
1575: i++;
1576: } else if (i+bs > n) { /* complete block doesn't exist (at idx end) */
1577: sizes[j] = 1; /* Also makes sure atleast 'bs' values exist for next else */
1578: i++;
1579: } else { /* Begining of the block, so check if the complete block exists */
1580: flg = PETSC_TRUE;
1581: for (k=1; k<bs; k++) {
1582: if (row+k != idx[i+k]) { /* break in the block */
1583: flg = PETSC_FALSE;
1584: break;
1585: }
1586: }
1587: if (flg) { /* No break in the bs */
1588: sizes[j] = bs;
1589: i+= bs;
1590: } else {
1591: sizes[j] = 1;
1592: i++;
1593: }
1594: }
1595: }
1596: *bs_max = j;
1597: return(0);
1598: }
1599:
1602: PetscErrorCode MatZeroRows_SeqBAIJ(Mat A,PetscInt is_n,const PetscInt is_idx[],PetscScalar diag)
1603: {
1604: Mat_SeqBAIJ *baij=(Mat_SeqBAIJ*)A->data;
1606: PetscInt i,j,k,count,*rows;
1607: PetscInt bs=A->bs,bs2=baij->bs2,*sizes,row,bs_max;
1608: PetscScalar zero = 0.0;
1609: MatScalar *aa;
1612: /* Make a copy of the IS and sort it */
1613: /* allocate memory for rows,sizes */
1614: PetscMalloc((3*is_n+1)*sizeof(PetscInt),&rows);
1615: sizes = rows + is_n;
1617: /* copy IS values to rows, and sort them */
1618: for (i=0; i<is_n; i++) { rows[i] = is_idx[i]; }
1619: PetscSortInt(is_n,rows);
1620: if (baij->keepzeroedrows) {
1621: for (i=0; i<is_n; i++) { sizes[i] = 1; }
1622: bs_max = is_n;
1623: A->same_nonzero = PETSC_TRUE;
1624: } else {
1625: MatZeroRows_SeqBAIJ_Check_Blocks(rows,is_n,bs,sizes,&bs_max);
1626: A->same_nonzero = PETSC_FALSE;
1627: }
1629: for (i=0,j=0; i<bs_max; j+=sizes[i],i++) {
1630: row = rows[j];
1631: if (row < 0 || row > A->m) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"row %D out of range",row);
1632: count = (baij->i[row/bs +1] - baij->i[row/bs])*bs;
1633: aa = baij->a + baij->i[row/bs]*bs2 + (row%bs);
1634: if (sizes[i] == bs && !baij->keepzeroedrows) {
1635: if (diag != 0.0) {
1636: if (baij->ilen[row/bs] > 0) {
1637: baij->ilen[row/bs] = 1;
1638: baij->j[baij->i[row/bs]] = row/bs;
1639: PetscMemzero(aa,count*bs*sizeof(MatScalar));
1640: }
1641: /* Now insert all the diagonal values for this bs */
1642: for (k=0; k<bs; k++) {
1643: (*A->ops->setvalues)(A,1,rows+j+k,1,rows+j+k,&diag,INSERT_VALUES);
1644: }
1645: } else { /* (diag == 0.0) */
1646: baij->ilen[row/bs] = 0;
1647: } /* end (diag == 0.0) */
1648: } else { /* (sizes[i] != bs) */
1649: #if defined (PETSC_USE_DEBUG)
1650: if (sizes[i] != 1) SETERRQ(PETSC_ERR_PLIB,"Internal Error. Value should be 1");
1651: #endif
1652: for (k=0; k<count; k++) {
1653: aa[0] = zero;
1654: aa += bs;
1655: }
1656: if (diag != 0.0) {
1657: (*A->ops->setvalues)(A,1,rows+j,1,rows+j,&diag,INSERT_VALUES);
1658: }
1659: }
1660: }
1662: PetscFree(rows);
1663: MatAssemblyEnd_SeqBAIJ(A,MAT_FINAL_ASSEMBLY);
1664: return(0);
1665: }
1669: PetscErrorCode MatSetValues_SeqBAIJ(Mat A,PetscInt m,const PetscInt im[],PetscInt n,const PetscInt in[],const PetscScalar v[],InsertMode is)
1670: {
1671: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
1672: PetscInt *rp,k,low,high,t,ii,row,nrow,i,col,l,rmax,N,lastcol = -1;
1673: PetscInt *imax=a->imax,*ai=a->i,*ailen=a->ilen;
1674: PetscInt *aj=a->j,nonew=a->nonew,bs=A->bs,brow,bcol;
1676: PetscInt ridx,cidx,bs2=a->bs2;
1677: PetscTruth roworiented=a->roworiented;
1678: MatScalar *ap,value,*aa=a->a,*bap;
1681: for (k=0; k<m; k++) { /* loop over added rows */
1682: row = im[k];
1683: brow = row/bs;
1684: if (row < 0) continue;
1685: #if defined(PETSC_USE_DEBUG)
1686: if (row >= A->m) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Row too large: row %D max %D",row,A->m-1);
1687: #endif
1688: rp = aj + ai[brow];
1689: ap = aa + bs2*ai[brow];
1690: rmax = imax[brow];
1691: nrow = ailen[brow];
1692: low = 0;
1693: high = nrow;
1694: for (l=0; l<n; l++) { /* loop over added columns */
1695: if (in[l] < 0) continue;
1696: #if defined(PETSC_USE_DEBUG)
1697: if (in[l] >= A->n) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Column too large: col %D max %D",in[l],A->n-1);
1698: #endif
1699: col = in[l]; bcol = col/bs;
1700: ridx = row % bs; cidx = col % bs;
1701: if (roworiented) {
1702: value = v[l + k*n];
1703: } else {
1704: value = v[k + l*m];
1705: }
1706: if (col <= lastcol) low = 0; else high = nrow;
1707: lastcol = col;
1708: while (high-low > 7) {
1709: t = (low+high)/2;
1710: if (rp[t] > bcol) high = t;
1711: else low = t;
1712: }
1713: for (i=low; i<high; i++) {
1714: if (rp[i] > bcol) break;
1715: if (rp[i] == bcol) {
1716: bap = ap + bs2*i + bs*cidx + ridx;
1717: if (is == ADD_VALUES) *bap += value;
1718: else *bap = value;
1719: goto noinsert1;
1720: }
1721: }
1722: if (nonew == 1) goto noinsert1;
1723: if (nonew == -1) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"Inserting a new nonzero (%D, %D) in the matrix", row, col);
1724: MatSeqXAIJReallocateAIJ(a,bs2,nrow,brow,bcol,rmax,aa,ai,aj,a->mbs,rp,ap,imax,nonew);
1725: N = nrow++ - 1;
1726: /* shift up all the later entries in this row */
1727: for (ii=N; ii>=i; ii--) {
1728: rp[ii+1] = rp[ii];
1729: PetscMemcpy(ap+bs2*(ii+1),ap+bs2*(ii),bs2*sizeof(MatScalar));
1730: }
1731: if (N>=i) {
1732: PetscMemzero(ap+bs2*i,bs2*sizeof(MatScalar));
1733: }
1734: rp[i] = bcol;
1735: ap[bs2*i + bs*cidx + ridx] = value;
1736: a->nz++;
1737: noinsert1:;
1738: low = i;
1739: }
1740: ailen[brow] = nrow;
1741: }
1742: A->same_nonzero = PETSC_FALSE;
1743: return(0);
1744: }
1749: PetscErrorCode MatILUFactor_SeqBAIJ(Mat inA,IS row,IS col,MatFactorInfo *info)
1750: {
1751: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)inA->data;
1752: Mat outA;
1754: PetscTruth row_identity,col_identity;
1757: if (info->levels != 0) SETERRQ(PETSC_ERR_SUP,"Only levels = 0 supported for in-place ILU");
1758: ISIdentity(row,&row_identity);
1759: ISIdentity(col,&col_identity);
1760: if (!row_identity || !col_identity) {
1761: SETERRQ(PETSC_ERR_ARG_WRONG,"Row and column permutations must be identity for in-place ILU");
1762: }
1764: outA = inA;
1765: inA->factor = FACTOR_LU;
1767: if (!a->diag) {
1768: MatMarkDiagonal_SeqBAIJ(inA);
1769: }
1771: a->row = row;
1772: a->col = col;
1773: PetscObjectReference((PetscObject)row);
1774: PetscObjectReference((PetscObject)col);
1775:
1776: /* Create the invert permutation so that it can be used in MatLUFactorNumeric() */
1777: ISInvertPermutation(col,PETSC_DECIDE,&a->icol);
1778: PetscLogObjectParent(inA,a->icol);
1779:
1780: /*
1781: Blocksize 2, 3, 4, 5, 6 and 7 have a special faster factorization/solver
1782: for ILU(0) factorization with natural ordering
1783: */
1784: if (inA->bs < 8) {
1785: MatSeqBAIJ_UpdateFactorNumeric_NaturalOrdering(inA);
1786: } else {
1787: if (!a->solve_work) {
1788: PetscMalloc((inA->m+inA->bs)*sizeof(PetscScalar),&a->solve_work);
1789: PetscLogObjectMemory(inA,(inA->m+inA->bs)*sizeof(PetscScalar));
1790: }
1791: }
1793: MatLUFactorNumeric(inA,info,&outA);
1795: return(0);
1796: }
1799: PetscErrorCode MatPrintHelp_SeqBAIJ(Mat A)
1800: {
1801: static PetscTruth called = PETSC_FALSE;
1802: MPI_Comm comm = A->comm;
1803: PetscErrorCode ierr;
1806: if (called) {return(0);} else called = PETSC_TRUE;
1807: (*PetscHelpPrintf)(comm," Options for MATSEQBAIJ and MATMPIBAIJ matrix formats (the defaults):\n");
1808: (*PetscHelpPrintf)(comm," -mat_block_size <block_size>\n");
1809: return(0);
1810: }
1815: PetscErrorCode PETSCMAT_DLLEXPORT MatSeqBAIJSetColumnIndices_SeqBAIJ(Mat mat,PetscInt *indices)
1816: {
1817: Mat_SeqBAIJ *baij = (Mat_SeqBAIJ *)mat->data;
1818: PetscInt i,nz,nbs;
1821: nz = baij->maxnz/baij->bs2;
1822: nbs = baij->nbs;
1823: for (i=0; i<nz; i++) {
1824: baij->j[i] = indices[i];
1825: }
1826: baij->nz = nz;
1827: for (i=0; i<nbs; i++) {
1828: baij->ilen[i] = baij->imax[i];
1829: }
1831: return(0);
1832: }
1837: /*@
1838: MatSeqBAIJSetColumnIndices - Set the column indices for all the rows
1839: in the matrix.
1841: Input Parameters:
1842: + mat - the SeqBAIJ matrix
1843: - indices - the column indices
1845: Level: advanced
1847: Notes:
1848: This can be called if you have precomputed the nonzero structure of the
1849: matrix and want to provide it to the matrix object to improve the performance
1850: of the MatSetValues() operation.
1852: You MUST have set the correct numbers of nonzeros per row in the call to
1853: MatCreateSeqBAIJ(), and the columns indices MUST be sorted.
1855: MUST be called before any calls to MatSetValues();
1857: @*/
1858: PetscErrorCode PETSCMAT_DLLEXPORT MatSeqBAIJSetColumnIndices(Mat mat,PetscInt *indices)
1859: {
1860: PetscErrorCode ierr,(*f)(Mat,PetscInt *);
1865: PetscObjectQueryFunction((PetscObject)mat,"MatSeqBAIJSetColumnIndices_C",(void (**)(void))&f);
1866: if (f) {
1867: (*f)(mat,indices);
1868: } else {
1869: SETERRQ(PETSC_ERR_ARG_WRONG,"Wrong type of matrix to set column indices");
1870: }
1871: return(0);
1872: }
1876: PetscErrorCode MatGetRowMax_SeqBAIJ(Mat A,Vec v)
1877: {
1878: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
1880: PetscInt i,j,n,row,bs,*ai,*aj,mbs;
1881: PetscReal atmp;
1882: PetscScalar *x,zero = 0.0;
1883: MatScalar *aa;
1884: PetscInt ncols,brow,krow,kcol;
1887: if (A->factor) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Not for factored matrix");
1888: bs = A->bs;
1889: aa = a->a;
1890: ai = a->i;
1891: aj = a->j;
1892: mbs = a->mbs;
1894: VecSet(v,zero);
1895: VecGetArray(v,&x);
1896: VecGetLocalSize(v,&n);
1897: if (n != A->m) SETERRQ(PETSC_ERR_ARG_SIZ,"Nonconforming matrix and vector");
1898: for (i=0; i<mbs; i++) {
1899: ncols = ai[1] - ai[0]; ai++;
1900: brow = bs*i;
1901: for (j=0; j<ncols; j++){
1902: /* bcol = bs*(*aj); */
1903: for (kcol=0; kcol<bs; kcol++){
1904: for (krow=0; krow<bs; krow++){
1905: atmp = PetscAbsScalar(*aa); aa++;
1906: row = brow + krow; /* row index */
1907: /* printf("val[%d,%d]: %g\n",row,bcol+kcol,atmp); */
1908: if (PetscAbsScalar(x[row]) < atmp) x[row] = atmp;
1909: }
1910: }
1911: aj++;
1912: }
1913: }
1914: VecRestoreArray(v,&x);
1915: return(0);
1916: }
1920: PetscErrorCode MatSetUpPreallocation_SeqBAIJ(Mat A)
1921: {
1925: MatSeqBAIJSetPreallocation_SeqBAIJ(A,1,PETSC_DEFAULT,0);
1926: return(0);
1927: }
1931: PetscErrorCode MatGetArray_SeqBAIJ(Mat A,PetscScalar *array[])
1932: {
1933: Mat_SeqBAIJ *a = (Mat_SeqBAIJ*)A->data;
1935: *array = a->a;
1936: return(0);
1937: }
1941: PetscErrorCode MatRestoreArray_SeqBAIJ(Mat A,PetscScalar *array[])
1942: {
1944: return(0);
1945: }
1947: #include petscblaslapack.h
1950: PetscErrorCode MatAXPY_SeqBAIJ(Mat Y,PetscScalar a,Mat X,MatStructure str)
1951: {
1952: Mat_SeqBAIJ *x = (Mat_SeqBAIJ *)X->data,*y = (Mat_SeqBAIJ *)Y->data;
1954: PetscInt i,bs=Y->bs,j,bs2;
1955: PetscBLASInt one=1,bnz = (PetscBLASInt)x->nz;
1958: if (str == SAME_NONZERO_PATTERN) {
1959: PetscScalar alpha = a;
1960: BLASaxpy_(&bnz,&alpha,x->a,&one,y->a,&one);
1961: } else if (str == SUBSET_NONZERO_PATTERN) { /* nonzeros of X is a subset of Y's */
1962: if (y->xtoy && y->XtoY != X) {
1963: PetscFree(y->xtoy);
1964: MatDestroy(y->XtoY);
1965: }
1966: if (!y->xtoy) { /* get xtoy */
1967: MatAXPYGetxtoy_Private(x->mbs,x->i,x->j,PETSC_NULL, y->i,y->j,PETSC_NULL, &y->xtoy);
1968: y->XtoY = X;
1969: }
1970: bs2 = bs*bs;
1971: for (i=0; i<x->nz; i++) {
1972: j = 0;
1973: while (j < bs2){
1974: y->a[bs2*y->xtoy[i]+j] += a*(x->a[bs2*i+j]);
1975: j++;
1976: }
1977: }
1978: PetscLogInfo((0,"MatAXPY_SeqBAIJ: ratio of nnz(X)/nnz(Y): %D/%D = %g\n",bs2*x->nz,bs2*y->nz,(PetscReal)(bs2*x->nz)/(bs2*y->nz)));
1979: } else {
1980: MatAXPY_Basic(Y,a,X,str);
1981: }
1982: return(0);
1983: }
1985: /* -------------------------------------------------------------------*/
1986: static struct _MatOps MatOps_Values = {MatSetValues_SeqBAIJ,
1987: MatGetRow_SeqBAIJ,
1988: MatRestoreRow_SeqBAIJ,
1989: MatMult_SeqBAIJ_N,
1990: /* 4*/ MatMultAdd_SeqBAIJ_N,
1991: MatMultTranspose_SeqBAIJ,
1992: MatMultTransposeAdd_SeqBAIJ,
1993: MatSolve_SeqBAIJ_N,
1994: 0,
1995: 0,
1996: /*10*/ 0,
1997: MatLUFactor_SeqBAIJ,
1998: 0,
1999: 0,
2000: MatTranspose_SeqBAIJ,
2001: /*15*/ MatGetInfo_SeqBAIJ,
2002: MatEqual_SeqBAIJ,
2003: MatGetDiagonal_SeqBAIJ,
2004: MatDiagonalScale_SeqBAIJ,
2005: MatNorm_SeqBAIJ,
2006: /*20*/ 0,
2007: MatAssemblyEnd_SeqBAIJ,
2008: 0,
2009: MatSetOption_SeqBAIJ,
2010: MatZeroEntries_SeqBAIJ,
2011: /*25*/ MatZeroRows_SeqBAIJ,
2012: MatLUFactorSymbolic_SeqBAIJ,
2013: MatLUFactorNumeric_SeqBAIJ_N,
2014: MatCholeskyFactorSymbolic_SeqBAIJ,
2015: MatCholeskyFactorNumeric_SeqBAIJ_N,
2016: /*30*/ MatSetUpPreallocation_SeqBAIJ,
2017: MatILUFactorSymbolic_SeqBAIJ,
2018: MatICCFactorSymbolic_SeqBAIJ,
2019: MatGetArray_SeqBAIJ,
2020: MatRestoreArray_SeqBAIJ,
2021: /*35*/ MatDuplicate_SeqBAIJ,
2022: 0,
2023: 0,
2024: MatILUFactor_SeqBAIJ,
2025: 0,
2026: /*40*/ MatAXPY_SeqBAIJ,
2027: MatGetSubMatrices_SeqBAIJ,
2028: MatIncreaseOverlap_SeqBAIJ,
2029: MatGetValues_SeqBAIJ,
2030: 0,
2031: /*45*/ MatPrintHelp_SeqBAIJ,
2032: MatScale_SeqBAIJ,
2033: 0,
2034: 0,
2035: 0,
2036: /*50*/ 0,
2037: MatGetRowIJ_SeqBAIJ,
2038: MatRestoreRowIJ_SeqBAIJ,
2039: 0,
2040: 0,
2041: /*55*/ 0,
2042: 0,
2043: 0,
2044: 0,
2045: MatSetValuesBlocked_SeqBAIJ,
2046: /*60*/ MatGetSubMatrix_SeqBAIJ,
2047: MatDestroy_SeqBAIJ,
2048: MatView_SeqBAIJ,
2049: MatGetPetscMaps_Petsc,
2050: 0,
2051: /*65*/ 0,
2052: 0,
2053: 0,
2054: 0,
2055: 0,
2056: /*70*/ MatGetRowMax_SeqBAIJ,
2057: MatConvert_Basic,
2058: 0,
2059: 0,
2060: 0,
2061: /*75*/ 0,
2062: 0,
2063: 0,
2064: 0,
2065: 0,
2066: /*80*/ 0,
2067: 0,
2068: 0,
2069: 0,
2070: MatLoad_SeqBAIJ,
2071: /*85*/ 0,
2072: 0,
2073: 0,
2074: 0,
2075: 0,
2076: /*90*/ 0,
2077: 0,
2078: 0,
2079: 0,
2080: 0,
2081: /*95*/ 0,
2082: 0,
2083: 0,
2084: 0};
2089: PetscErrorCode PETSCMAT_DLLEXPORT MatStoreValues_SeqBAIJ(Mat mat)
2090: {
2091: Mat_SeqBAIJ *aij = (Mat_SeqBAIJ *)mat->data;
2092: PetscInt nz = aij->i[mat->m]*mat->bs*aij->bs2;
2096: if (aij->nonew != 1) {
2097: SETERRQ(PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NO_NEW_NONZERO_LOCATIONS);first");
2098: }
2100: /* allocate space for values if not already there */
2101: if (!aij->saved_values) {
2102: PetscMalloc((nz+1)*sizeof(PetscScalar),&aij->saved_values);
2103: }
2105: /* copy values over */
2106: PetscMemcpy(aij->saved_values,aij->a,nz*sizeof(PetscScalar));
2107: return(0);
2108: }
2114: PetscErrorCode PETSCMAT_DLLEXPORT MatRetrieveValues_SeqBAIJ(Mat mat)
2115: {
2116: Mat_SeqBAIJ *aij = (Mat_SeqBAIJ *)mat->data;
2118: PetscInt nz = aij->i[mat->m]*mat->bs*aij->bs2;
2121: if (aij->nonew != 1) {
2122: SETERRQ(PETSC_ERR_ORDER,"Must call MatSetOption(A,MAT_NO_NEW_NONZERO_LOCATIONS);first");
2123: }
2124: if (!aij->saved_values) {
2125: SETERRQ(PETSC_ERR_ORDER,"Must call MatStoreValues(A);first");
2126: }
2128: /* copy values over */
2129: PetscMemcpy(aij->a,aij->saved_values,nz*sizeof(PetscScalar));
2130: return(0);
2131: }
2142: PetscErrorCode PETSCMAT_DLLEXPORT MatSeqBAIJSetPreallocation_SeqBAIJ(Mat B,PetscInt bs,PetscInt nz,PetscInt *nnz)
2143: {
2144: Mat_SeqBAIJ *b;
2146: PetscInt i,mbs,nbs,bs2,newbs = bs;
2147: PetscTruth flg,skipallocation = PETSC_FALSE;
2151: if (nz == MAT_SKIP_ALLOCATION) {
2152: skipallocation = PETSC_TRUE;
2153: nz = 0;
2154: }
2156: B->preallocated = PETSC_TRUE;
2157: PetscOptionsGetInt(B->prefix,"-mat_block_size",&newbs,PETSC_NULL);
2158: if (nnz && newbs != bs) {
2159: SETERRQ(PETSC_ERR_ARG_WRONG,"Cannot change blocksize from command line if setting nnz");
2160: }
2161: bs = newbs;
2163: mbs = B->m/bs;
2164: nbs = B->n/bs;
2165: bs2 = bs*bs;
2167: if (mbs*bs!=B->m || nbs*bs!=B->n) {
2168: SETERRQ3(PETSC_ERR_ARG_SIZ,"Number rows %D, cols %D must be divisible by blocksize %D",B->m,B->n,bs);
2169: }
2171: if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 5;
2172: if (nz < 0) SETERRQ1(PETSC_ERR_ARG_OUTOFRANGE,"nz cannot be less than 0: value %D",nz);
2173: if (nnz) {
2174: for (i=0; i<mbs; i++) {
2175: if (nnz[i] < 0) SETERRQ2(PETSC_ERR_ARG_OUTOFRANGE,"nnz cannot be less than 0: local row %D value %D",i,nnz[i]);
2176: if (nnz[i] > nbs) SETERRQ3(PETSC_ERR_ARG_OUTOFRANGE,"nnz cannot be greater than block row length: local row %D value %D rowlength %D",i,nnz[i],nbs);
2177: }
2178: }
2180: b = (Mat_SeqBAIJ*)B->data;
2181: PetscOptionsHasName(PETSC_NULL,"-mat_no_unroll",&flg);
2182: B->ops->solve = MatSolve_SeqBAIJ_Update;
2183: B->ops->solvetranspose = MatSolveTranspose_SeqBAIJ_Update;
2184: if (!flg) {
2185: switch (bs) {
2186: case 1:
2187: B->ops->lufactornumeric = MatLUFactorNumeric_SeqBAIJ_1;
2188: B->ops->mult = MatMult_SeqBAIJ_1;
2189: B->ops->multadd = MatMultAdd_SeqBAIJ_1;
2190: break;
2191: case 2:
2192: B->ops->lufactornumeric = MatLUFactorNumeric_SeqBAIJ_2;
2193: B->ops->mult = MatMult_SeqBAIJ_2;
2194: B->ops->multadd = MatMultAdd_SeqBAIJ_2;
2195: B->ops->pbrelax = MatPBRelax_SeqBAIJ_2;
2196: break;
2197: case 3:
2198: B->ops->lufactornumeric = MatLUFactorNumeric_SeqBAIJ_3;
2199: B->ops->mult = MatMult_SeqBAIJ_3;
2200: B->ops->multadd = MatMultAdd_SeqBAIJ_3;
2201: B->ops->pbrelax = MatPBRelax_SeqBAIJ_3;
2202: break;
2203: case 4:
2204: B->ops->lufactornumeric = MatLUFactorNumeric_SeqBAIJ_4;
2205: B->ops->mult = MatMult_SeqBAIJ_4;
2206: B->ops->multadd = MatMultAdd_SeqBAIJ_4;
2207: B->ops->pbrelax = MatPBRelax_SeqBAIJ_4;
2208: break;
2209: case 5:
2210: B->ops->lufactornumeric = MatLUFactorNumeric_SeqBAIJ_5;
2211: B->ops->mult = MatMult_SeqBAIJ_5;
2212: B->ops->multadd = MatMultAdd_SeqBAIJ_5;
2213: B->ops->pbrelax = MatPBRelax_SeqBAIJ_5;
2214: break;
2215: case 6:
2216: B->ops->lufactornumeric = MatLUFactorNumeric_SeqBAIJ_6;
2217: B->ops->mult = MatMult_SeqBAIJ_6;
2218: B->ops->multadd = MatMultAdd_SeqBAIJ_6;
2219: break;
2220: case 7:
2221: B->ops->lufactornumeric = MatLUFactorNumeric_SeqBAIJ_7;
2222: B->ops->mult = MatMult_SeqBAIJ_7;
2223: B->ops->multadd = MatMultAdd_SeqBAIJ_7;
2224: break;
2225: default:
2226: B->ops->lufactornumeric = MatLUFactorNumeric_SeqBAIJ_N;
2227: B->ops->mult = MatMult_SeqBAIJ_N;
2228: B->ops->multadd = MatMultAdd_SeqBAIJ_N;
2229: break;
2230: }
2231: }
2232: B->bs = bs;
2233: b->mbs = mbs;
2234: b->nbs = nbs;
2235: if (!skipallocation) {
2236: PetscMalloc2(mbs,PetscInt,&b->imax,mbs,PetscInt,&b->ilen);
2237: /* b->ilen will count nonzeros in each block row so far. */
2238: for (i=0; i<mbs; i++) { b->ilen[i] = 0;}
2239: if (!nnz) {
2240: if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 5;
2241: else if (nz <= 0) nz = 1;
2242: for (i=0; i<mbs; i++) b->imax[i] = nz;
2243: nz = nz*mbs;
2244: } else {
2245: nz = 0;
2246: for (i=0; i<mbs; i++) {b->imax[i] = nnz[i]; nz += nnz[i];}
2247: }
2249: /* allocate the matrix space */
2250: PetscMalloc3(bs2*nz,PetscScalar,&b->a,nz,PetscInt,&b->j,B->m+1,PetscInt,&b->i);
2251: PetscMemzero(b->a,nz*bs2*sizeof(MatScalar));
2252: PetscMemzero(b->j,nz*sizeof(PetscInt));
2253: b->singlemalloc = PETSC_TRUE;
2255: b->i[0] = 0;
2256: for (i=1; i<mbs+1; i++) {
2257: b->i[i] = b->i[i-1] + b->imax[i-1];
2258: }
2259: }
2261: B->bs = bs;
2262: b->bs2 = bs2;
2263: b->mbs = mbs;
2264: b->nz = 0;
2265: b->maxnz = nz*bs2;
2266: B->info.nz_unneeded = (PetscReal)b->maxnz;
2267: return(0);
2268: }
2271: /*MC
2272: MATSEQBAIJ - MATSEQBAIJ = "seqbaij" - A matrix type to be used for sequential block sparse matrices, based on
2273: block sparse compressed row format.
2275: Options Database Keys:
2276: . -mat_type seqbaij - sets the matrix type to "seqbaij" during a call to MatSetFromOptions()
2278: Level: beginner
2280: .seealso: MatCreateSeqBAIJ
2281: M*/
2286: PetscErrorCode PETSCMAT_DLLEXPORT MatCreate_SeqBAIJ(Mat B)
2287: {
2289: PetscMPIInt size;
2290: Mat_SeqBAIJ *b;
2293: MPI_Comm_size(B->comm,&size);
2294: if (size > 1) SETERRQ(PETSC_ERR_ARG_WRONG,"Comm must be of size 1");
2296: B->m = B->M = PetscMax(B->m,B->M);
2297: B->n = B->N = PetscMax(B->n,B->N);
2298: PetscNew(Mat_SeqBAIJ,&b);
2299: B->data = (void*)b;
2300: PetscMemcpy(B->ops,&MatOps_Values,sizeof(struct _MatOps));
2301: B->factor = 0;
2302: B->mapping = 0;
2303: b->row = 0;
2304: b->col = 0;
2305: b->icol = 0;
2306: b->reallocs = 0;
2307: b->saved_values = 0;
2308: #if defined(PETSC_USE_MAT_SINGLE)
2309: b->setvalueslen = 0;
2310: b->setvaluescopy = PETSC_NULL;
2311: #endif
2313: PetscMapCreateMPI(B->comm,B->m,B->m,&B->rmap);
2314: PetscMapCreateMPI(B->comm,B->n,B->n,&B->cmap);
2316: b->sorted = PETSC_FALSE;
2317: b->roworiented = PETSC_TRUE;
2318: b->nonew = 0;
2319: b->diag = 0;
2320: b->solve_work = 0;
2321: b->mult_work = 0;
2322: B->spptr = 0;
2323: B->info.nz_unneeded = (PetscReal)b->maxnz;
2324: b->keepzeroedrows = PETSC_FALSE;
2325: b->xtoy = 0;
2326: b->XtoY = 0;
2327: b->compressedrow.use = PETSC_FALSE;
2328: b->compressedrow.nrows = 0;
2329: b->compressedrow.i = PETSC_NULL;
2330: b->compressedrow.rindex = PETSC_NULL;
2331: b->compressedrow.checked = PETSC_FALSE;
2332: B->same_nonzero = PETSC_FALSE;
2334: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatSeqBAIJInvertBlockDiagonal_C",
2335: "MatInvertBlockDiagonal_SeqBAIJ",
2336: MatInvertBlockDiagonal_SeqBAIJ);
2337: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatStoreValues_C",
2338: "MatStoreValues_SeqBAIJ",
2339: MatStoreValues_SeqBAIJ);
2340: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatRetrieveValues_C",
2341: "MatRetrieveValues_SeqBAIJ",
2342: MatRetrieveValues_SeqBAIJ);
2343: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatSeqBAIJSetColumnIndices_C",
2344: "MatSeqBAIJSetColumnIndices_SeqBAIJ",
2345: MatSeqBAIJSetColumnIndices_SeqBAIJ);
2346: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatConvert_seqbaij_seqaij_C",
2347: "MatConvert_SeqBAIJ_SeqAIJ",
2348: MatConvert_SeqBAIJ_SeqAIJ);
2349: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatConvert_seqbaij_seqsbaij_C",
2350: "MatConvert_SeqBAIJ_SeqSBAIJ",
2351: MatConvert_SeqBAIJ_SeqSBAIJ);
2352: PetscObjectComposeFunctionDynamic((PetscObject)B,"MatSeqBAIJSetPreallocation_C",
2353: "MatSeqBAIJSetPreallocation_SeqBAIJ",
2354: MatSeqBAIJSetPreallocation_SeqBAIJ);
2355: return(0);
2356: }
2361: PetscErrorCode MatDuplicate_SeqBAIJ(Mat A,MatDuplicateOption cpvalues,Mat *B)
2362: {
2363: Mat C;
2364: Mat_SeqBAIJ *c,*a = (Mat_SeqBAIJ*)A->data;
2366: PetscInt i,mbs = a->mbs,nz = a->nz,bs2 = a->bs2;
2369: if (a->i[mbs] != nz) SETERRQ(PETSC_ERR_PLIB,"Corrupt matrix");
2371: *B = 0;
2372: MatCreate(A->comm,&C);
2373: MatSetSizes(C,A->m,A->n,A->m,A->n);
2374: MatSetType(C,A->type_name);
2375: PetscMemcpy(C->ops,A->ops,sizeof(struct _MatOps));
2376: c = (Mat_SeqBAIJ*)C->data;
2378: C->M = A->M;
2379: C->N = A->N;
2380: C->bs = A->bs;
2381: c->bs2 = a->bs2;
2382: c->mbs = a->mbs;
2383: c->nbs = a->nbs;
2385: PetscMalloc2(mbs,PetscInt,&c->imax,mbs,PetscInt,&c->ilen);
2386: for (i=0; i<mbs; i++) {
2387: c->imax[i] = a->imax[i];
2388: c->ilen[i] = a->ilen[i];
2389: }
2391: /* allocate the matrix space */
2392: PetscMalloc3(bs2*nz,PetscScalar,&c->a,nz,PetscInt,&c->j,mbs+1,PetscInt,&c->i);
2393: c->singlemalloc = PETSC_TRUE;
2394: PetscMemcpy(c->i,a->i,(mbs+1)*sizeof(PetscInt));
2395: if (mbs > 0) {
2396: PetscMemcpy(c->j,a->j,nz*sizeof(PetscInt));
2397: if (cpvalues == MAT_COPY_VALUES) {
2398: PetscMemcpy(c->a,a->a,bs2*nz*sizeof(MatScalar));
2399: } else {
2400: PetscMemzero(c->a,bs2*nz*sizeof(MatScalar));
2401: }
2402: }
2403: c->sorted = a->sorted;
2404: c->roworiented = a->roworiented;
2405: c->nonew = a->nonew;
2407: if (a->diag) {
2408: PetscMalloc((mbs+1)*sizeof(PetscInt),&c->diag);
2409: PetscLogObjectMemory(C,(mbs+1)*sizeof(PetscInt));
2410: for (i=0; i<mbs; i++) {
2411: c->diag[i] = a->diag[i];
2412: }
2413: } else c->diag = 0;
2414: c->nz = a->nz;
2415: c->maxnz = a->maxnz;
2416: c->solve_work = 0;
2417: c->mult_work = 0;
2418: C->preallocated = PETSC_TRUE;
2419: C->assembled = PETSC_TRUE;
2421: c->compressedrow.use = a->compressedrow.use;
2422: c->compressedrow.nrows = a->compressedrow.nrows;
2423: c->compressedrow.checked = a->compressedrow.checked;
2424: if ( a->compressedrow.checked && a->compressedrow.use){
2425: i = a->compressedrow.nrows;
2426: PetscMalloc((2*i+1)*sizeof(PetscInt),&c->compressedrow.i);
2427: c->compressedrow.rindex = c->compressedrow.i + i + 1;
2428: PetscMemcpy(c->compressedrow.i,a->compressedrow.i,(i+1)*sizeof(PetscInt));
2429: PetscMemcpy(c->compressedrow.rindex,a->compressedrow.rindex,i*sizeof(PetscInt));
2430: } else {
2431: c->compressedrow.use = PETSC_FALSE;
2432: c->compressedrow.i = PETSC_NULL;
2433: c->compressedrow.rindex = PETSC_NULL;
2434: }
2435: C->same_nonzero = A->same_nonzero;
2436: *B = C;
2437: PetscFListDuplicate(A->qlist,&C->qlist);
2438: return(0);
2439: }
2443: PetscErrorCode MatLoad_SeqBAIJ(PetscViewer viewer, MatType type,Mat *A)
2444: {
2445: Mat_SeqBAIJ *a;
2446: Mat B;
2448: PetscInt i,nz,header[4],*rowlengths=0,M,N,bs=1;
2449: PetscInt *mask,mbs,*jj,j,rowcount,nzcount,k,*browlengths,maskcount;
2450: PetscInt kmax,jcount,block,idx,point,nzcountb,extra_rows;
2451: PetscInt *masked,nmask,tmp,bs2,ishift;
2452: PetscMPIInt size;
2453: int fd;
2454: PetscScalar *aa;
2455: MPI_Comm comm = ((PetscObject)viewer)->comm;
2458: PetscOptionsGetInt(PETSC_NULL,"-matload_block_size",&bs,PETSC_NULL);
2459: bs2 = bs*bs;
2461: MPI_Comm_size(comm,&size);
2462: if (size > 1) SETERRQ(PETSC_ERR_ARG_WRONG,"view must have one processor");
2463: PetscViewerBinaryGetDescriptor(viewer,&fd);
2464: PetscBinaryRead(fd,header,4,PETSC_INT);
2465: if (header[0] != MAT_FILE_COOKIE) SETERRQ(PETSC_ERR_FILE_UNEXPECTED,"not Mat object");
2466: M = header[1]; N = header[2]; nz = header[3];
2468: if (header[3] < 0) {
2469: SETERRQ(PETSC_ERR_FILE_UNEXPECTED,"Matrix stored in special format, cannot load as SeqBAIJ");
2470: }
2472: if (M != N) SETERRQ(PETSC_ERR_SUP,"Can only do square matrices");
2474: /*
2475: This code adds extra rows to make sure the number of rows is
2476: divisible by the blocksize
2477: */
2478: mbs = M/bs;
2479: extra_rows = bs - M + bs*(mbs);
2480: if (extra_rows == bs) extra_rows = 0;
2481: else mbs++;
2482: if (extra_rows) {
2483: PetscLogInfo((0,"MatLoad_SeqBAIJ:Padding loaded matrix to match blocksize\n"));
2484: }
2486: /* read in row lengths */
2487: PetscMalloc((M+extra_rows)*sizeof(PetscInt),&rowlengths);
2488: PetscBinaryRead(fd,rowlengths,M,PETSC_INT);
2489: for (i=0; i<extra_rows; i++) rowlengths[M+i] = 1;
2491: /* read in column indices */
2492: PetscMalloc((nz+extra_rows)*sizeof(PetscInt),&jj);
2493: PetscBinaryRead(fd,jj,nz,PETSC_INT);
2494: for (i=0; i<extra_rows; i++) jj[nz+i] = M+i;
2496: /* loop over row lengths determining block row lengths */
2497: PetscMalloc(mbs*sizeof(PetscInt),&browlengths);
2498: PetscMemzero(browlengths,mbs*sizeof(PetscInt));
2499: PetscMalloc(2*mbs*sizeof(PetscInt),&mask);
2500: PetscMemzero(mask,mbs*sizeof(PetscInt));
2501: masked = mask + mbs;
2502: rowcount = 0; nzcount = 0;
2503: for (i=0; i<mbs; i++) {
2504: nmask = 0;
2505: for (j=0; j<bs; j++) {
2506: kmax = rowlengths[rowcount];
2507: for (k=0; k<kmax; k++) {
2508: tmp = jj[nzcount++]/bs;
2509: if (!mask[tmp]) {masked[nmask++] = tmp; mask[tmp] = 1;}
2510: }
2511: rowcount++;
2512: }
2513: browlengths[i] += nmask;
2514: /* zero out the mask elements we set */
2515: for (j=0; j<nmask; j++) mask[masked[j]] = 0;
2516: }
2518: /* create our matrix */
2519: MatCreate(comm,&B);
2520: MatSetSizes(B,PETSC_DECIDE,PETSC_DECIDE,M+extra_rows,N+extra_rows);
2521: MatSetType(B,type);
2522: MatSeqBAIJSetPreallocation_SeqBAIJ(B,bs,0,browlengths);
2523: a = (Mat_SeqBAIJ*)B->data;
2525: /* set matrix "i" values */
2526: a->i[0] = 0;
2527: for (i=1; i<= mbs; i++) {
2528: a->i[i] = a->i[i-1] + browlengths[i-1];
2529: a->ilen[i-1] = browlengths[i-1];
2530: }
2531: a->nz = 0;
2532: for (i=0; i<mbs; i++) a->nz += browlengths[i];
2534: /* read in nonzero values */
2535: PetscMalloc((nz+extra_rows)*sizeof(PetscScalar),&aa);
2536: PetscBinaryRead(fd,aa,nz,PETSC_SCALAR);
2537: for (i=0; i<extra_rows; i++) aa[nz+i] = 1.0;
2539: /* set "a" and "j" values into matrix */
2540: nzcount = 0; jcount = 0;
2541: for (i=0; i<mbs; i++) {
2542: nzcountb = nzcount;
2543: nmask = 0;
2544: for (j=0; j<bs; j++) {
2545: kmax = rowlengths[i*bs+j];
2546: for (k=0; k<kmax; k++) {
2547: tmp = jj[nzcount++]/bs;
2548: if (!mask[tmp]) { masked[nmask++] = tmp; mask[tmp] = 1;}
2549: }
2550: }
2551: /* sort the masked values */
2552: PetscSortInt(nmask,masked);
2554: /* set "j" values into matrix */
2555: maskcount = 1;
2556: for (j=0; j<nmask; j++) {
2557: a->j[jcount++] = masked[j];
2558: mask[masked[j]] = maskcount++;
2559: }
2560: /* set "a" values into matrix */
2561: ishift = bs2*a->i[i];
2562: for (j=0; j<bs; j++) {
2563: kmax = rowlengths[i*bs+j];
2564: for (k=0; k<kmax; k++) {
2565: tmp = jj[nzcountb]/bs ;
2566: block = mask[tmp] - 1;
2567: point = jj[nzcountb] - bs*tmp;
2568: idx = ishift + bs2*block + j + bs*point;
2569: a->a[idx] = (MatScalar)aa[nzcountb++];
2570: }
2571: }
2572: /* zero out the mask elements we set */
2573: for (j=0; j<nmask; j++) mask[masked[j]] = 0;
2574: }
2575: if (jcount != a->nz) SETERRQ(PETSC_ERR_FILE_UNEXPECTED,"Bad binary matrix");
2577: PetscFree(rowlengths);
2578: PetscFree(browlengths);
2579: PetscFree(aa);
2580: PetscFree(jj);
2581: PetscFree(mask);
2583: MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);
2584: MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);
2585: MatView_Private(B);
2587: *A = B;
2588: return(0);
2589: }
2593: /*@C
2594: MatCreateSeqBAIJ - Creates a sparse matrix in block AIJ (block
2595: compressed row) format. For good matrix assembly performance the
2596: user should preallocate the matrix storage by setting the parameter nz
2597: (or the array nnz). By setting these parameters accurately, performance
2598: during matrix assembly can be increased by more than a factor of 50.
2600: Collective on MPI_Comm
2602: Input Parameters:
2603: + comm - MPI communicator, set to PETSC_COMM_SELF
2604: . bs - size of block
2605: . m - number of rows
2606: . n - number of columns
2607: . nz - number of nonzero blocks per block row (same for all rows)
2608: - nnz - array containing the number of nonzero blocks in the various block rows
2609: (possibly different for each block row) or PETSC_NULL
2611: Output Parameter:
2612: . A - the matrix
2614: Options Database Keys:
2615: . -mat_no_unroll - uses code that does not unroll the loops in the
2616: block calculations (much slower)
2617: . -mat_block_size - size of the blocks to use
2619: Level: intermediate
2621: Notes:
2622: The number of rows and columns must be divisible by blocksize.
2624: If the nnz parameter is given then the nz parameter is ignored
2626: A nonzero block is any block that as 1 or more nonzeros in it
2628: The block AIJ format is fully compatible with standard Fortran 77
2629: storage. That is, the stored row and column indices can begin at
2630: either one (as in Fortran) or zero. See the users' manual for details.
2632: Specify the preallocated storage with either nz or nnz (not both).
2633: Set nz=PETSC_DEFAULT and nnz=PETSC_NULL for PETSc to control dynamic memory
2634: allocation. For additional details, see the users manual chapter on
2635: matrices.
2637: .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatCreateMPIBAIJ()
2638: @*/
2639: PetscErrorCode PETSCMAT_DLLEXPORT MatCreateSeqBAIJ(MPI_Comm comm,PetscInt bs,PetscInt m,PetscInt n,PetscInt nz,const PetscInt nnz[],Mat *A)
2640: {
2642:
2644: MatCreate(comm,A);
2645: MatSetSizes(*A,m,n,m,n);
2646: MatSetType(*A,MATSEQBAIJ);
2647: MatSeqBAIJSetPreallocation_SeqBAIJ(*A,bs,nz,(PetscInt*)nnz);
2648: return(0);
2649: }
2653: /*@C
2654: MatSeqBAIJSetPreallocation - Sets the block size and expected nonzeros
2655: per row in the matrix. For good matrix assembly performance the
2656: user should preallocate the matrix storage by setting the parameter nz
2657: (or the array nnz). By setting these parameters accurately, performance
2658: during matrix assembly can be increased by more than a factor of 50.
2660: Collective on MPI_Comm
2662: Input Parameters:
2663: + A - the matrix
2664: . bs - size of block
2665: . nz - number of block nonzeros per block row (same for all rows)
2666: - nnz - array containing the number of block nonzeros in the various block rows
2667: (possibly different for each block row) or PETSC_NULL
2669: Options Database Keys:
2670: . -mat_no_unroll - uses code that does not unroll the loops in the
2671: block calculations (much slower)
2672: . -mat_block_size - size of the blocks to use
2674: Level: intermediate
2676: Notes:
2677: If the nnz parameter is given then the nz parameter is ignored
2679: The block AIJ format is fully compatible with standard Fortran 77
2680: storage. That is, the stored row and column indices can begin at
2681: either one (as in Fortran) or zero. See the users' manual for details.
2683: Specify the preallocated storage with either nz or nnz (not both).
2684: Set nz=PETSC_DEFAULT and nnz=PETSC_NULL for PETSc to control dynamic memory
2685: allocation. For additional details, see the users manual chapter on
2686: matrices.
2688: .seealso: MatCreate(), MatCreateSeqAIJ(), MatSetValues(), MatCreateMPIBAIJ()
2689: @*/
2690: PetscErrorCode PETSCMAT_DLLEXPORT MatSeqBAIJSetPreallocation(Mat B,PetscInt bs,PetscInt nz,const PetscInt nnz[])
2691: {
2692: PetscErrorCode ierr,(*f)(Mat,PetscInt,PetscInt,const PetscInt[]);
2695: PetscObjectQueryFunction((PetscObject)B,"MatSeqBAIJSetPreallocation_C",(void (**)(void))&f);
2696: if (f) {
2697: (*f)(B,bs,nz,nnz);
2698: }
2699: return(0);
2700: }