Actual source code: baij.c
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: /* defines MatSetValues_Seq_Hash(), MatAssemblyEnd_Seq_Hash(), MatSetUp_Seq_Hash() */
12: #define TYPE BAIJ
13: #define TYPE_BS
14: #include "../src/mat/impls/aij/seq/seqhashmatsetvalues.h"
15: #undef TYPE_BS
16: #define TYPE_BS _BS
17: #define TYPE_BS_ON
18: #include "../src/mat/impls/aij/seq/seqhashmatsetvalues.h"
19: #undef TYPE_BS
20: #include "../src/mat/impls/aij/seq/seqhashmat.h"
21: #undef TYPE
22: #undef TYPE_BS_ON
24: #if defined(PETSC_HAVE_HYPRE)
25: PETSC_INTERN PetscErrorCode MatConvert_AIJ_HYPRE(Mat, MatType, MatReuse, Mat *);
26: #endif
28: #if defined(PETSC_HAVE_MKL_SPARSE_OPTIMIZE)
29: PETSC_INTERN PetscErrorCode MatConvert_SeqBAIJ_SeqBAIJMKL(Mat, MatType, MatReuse, Mat *);
30: #endif
31: PETSC_INTERN PetscErrorCode MatConvert_XAIJ_IS(Mat, MatType, MatReuse, Mat *);
33: static PetscErrorCode MatGetColumnReductions_SeqBAIJ(Mat A, PetscInt type, PetscReal *reductions)
34: {
35: Mat_SeqBAIJ *a_aij = (Mat_SeqBAIJ *)A->data;
36: PetscInt m, n, ib, jb, bs = A->rmap->bs;
37: MatScalar *a_val = a_aij->a;
39: PetscFunctionBegin;
40: PetscCall(MatGetSize(A, &m, &n));
41: PetscCall(PetscArrayzero(reductions, n));
42: if (type == NORM_2) {
43: for (PetscInt i = a_aij->i[0]; i < a_aij->i[A->rmap->n / bs]; i++) {
44: for (jb = 0; jb < bs; jb++) {
45: for (ib = 0; ib < bs; ib++) {
46: reductions[A->cmap->rstart + a_aij->j[i] * bs + jb] += PetscAbsScalar(*a_val * *a_val);
47: a_val++;
48: }
49: }
50: }
51: } else if (type == NORM_1) {
52: for (PetscInt i = a_aij->i[0]; i < a_aij->i[A->rmap->n / bs]; i++) {
53: for (jb = 0; jb < bs; jb++) {
54: for (ib = 0; ib < bs; ib++) {
55: reductions[A->cmap->rstart + a_aij->j[i] * bs + jb] += PetscAbsScalar(*a_val);
56: a_val++;
57: }
58: }
59: }
60: } else if (type == NORM_INFINITY) {
61: for (PetscInt i = a_aij->i[0]; i < a_aij->i[A->rmap->n / bs]; i++) {
62: for (jb = 0; jb < bs; jb++) {
63: for (ib = 0; ib < bs; ib++) {
64: int col = A->cmap->rstart + a_aij->j[i] * bs + jb;
65: reductions[col] = PetscMax(PetscAbsScalar(*a_val), reductions[col]);
66: a_val++;
67: }
68: }
69: }
70: } else if (type == REDUCTION_SUM_REALPART || type == REDUCTION_MEAN_REALPART) {
71: for (PetscInt i = a_aij->i[0]; i < a_aij->i[A->rmap->n / bs]; i++) {
72: for (jb = 0; jb < bs; jb++) {
73: for (ib = 0; ib < bs; ib++) {
74: reductions[A->cmap->rstart + a_aij->j[i] * bs + jb] += PetscRealPart(*a_val);
75: a_val++;
76: }
77: }
78: }
79: } else if (type == REDUCTION_SUM_IMAGINARYPART || type == REDUCTION_MEAN_IMAGINARYPART) {
80: for (PetscInt i = a_aij->i[0]; i < a_aij->i[A->rmap->n / bs]; i++) {
81: for (jb = 0; jb < bs; jb++) {
82: for (ib = 0; ib < bs; ib++) {
83: reductions[A->cmap->rstart + a_aij->j[i] * bs + jb] += PetscImaginaryPart(*a_val);
84: a_val++;
85: }
86: }
87: }
88: } else SETERRQ(PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONG, "Unknown reduction type");
89: if (type == NORM_2) {
90: for (PetscInt i = 0; i < n; i++) reductions[i] = PetscSqrtReal(reductions[i]);
91: } else if (type == REDUCTION_MEAN_REALPART || type == REDUCTION_MEAN_IMAGINARYPART) {
92: for (PetscInt i = 0; i < n; i++) reductions[i] /= m;
93: }
94: PetscFunctionReturn(PETSC_SUCCESS);
95: }
97: PetscErrorCode MatInvertBlockDiagonal_SeqBAIJ(Mat A, const PetscScalar **values)
98: {
99: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
100: PetscInt *diag_offset, i, bs = A->rmap->bs, mbs = a->mbs, ipvt[5], bs2 = bs * bs, *v_pivots;
101: MatScalar *v = a->a, *odiag, *diag, work[25], *v_work;
102: PetscReal shift = 0.0;
103: PetscBool allowzeropivot, zeropivotdetected = PETSC_FALSE;
105: PetscFunctionBegin;
106: allowzeropivot = PetscNot(A->erroriffailure);
108: if (a->idiagvalid) {
109: if (values) *values = a->idiag;
110: PetscFunctionReturn(PETSC_SUCCESS);
111: }
112: PetscCall(MatMarkDiagonal_SeqBAIJ(A));
113: diag_offset = a->diag;
114: if (!a->idiag) { PetscCall(PetscMalloc1(bs2 * mbs, &a->idiag)); }
115: diag = a->idiag;
116: if (values) *values = a->idiag;
117: /* factor and invert each block */
118: switch (bs) {
119: case 1:
120: for (i = 0; i < mbs; i++) {
121: odiag = v + 1 * diag_offset[i];
122: diag[0] = odiag[0];
124: if (PetscAbsScalar(diag[0] + shift) < PETSC_MACHINE_EPSILON) {
125: if (allowzeropivot) {
126: A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
127: A->factorerror_zeropivot_value = PetscAbsScalar(diag[0]);
128: A->factorerror_zeropivot_row = i;
129: PetscCall(PetscInfo(A, "Zero pivot, row %" PetscInt_FMT "\n", i));
130: } else SETERRQ(PETSC_COMM_SELF, PETSC_ERR_MAT_LU_ZRPVT, "Zero pivot, row %" PetscInt_FMT " pivot value %g tolerance %g", i, (double)PetscAbsScalar(diag[0]), (double)PETSC_MACHINE_EPSILON);
131: }
133: diag[0] = (PetscScalar)1.0 / (diag[0] + shift);
134: diag += 1;
135: }
136: break;
137: case 2:
138: for (i = 0; i < mbs; i++) {
139: odiag = v + 4 * diag_offset[i];
140: diag[0] = odiag[0];
141: diag[1] = odiag[1];
142: diag[2] = odiag[2];
143: diag[3] = odiag[3];
144: PetscCall(PetscKernel_A_gets_inverse_A_2(diag, shift, allowzeropivot, &zeropivotdetected));
145: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
146: diag += 4;
147: }
148: break;
149: case 3:
150: for (i = 0; i < mbs; i++) {
151: odiag = v + 9 * diag_offset[i];
152: diag[0] = odiag[0];
153: diag[1] = odiag[1];
154: diag[2] = odiag[2];
155: diag[3] = odiag[3];
156: diag[4] = odiag[4];
157: diag[5] = odiag[5];
158: diag[6] = odiag[6];
159: diag[7] = odiag[7];
160: diag[8] = odiag[8];
161: PetscCall(PetscKernel_A_gets_inverse_A_3(diag, shift, allowzeropivot, &zeropivotdetected));
162: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
163: diag += 9;
164: }
165: break;
166: case 4:
167: for (i = 0; i < mbs; i++) {
168: odiag = v + 16 * diag_offset[i];
169: PetscCall(PetscArraycpy(diag, odiag, 16));
170: PetscCall(PetscKernel_A_gets_inverse_A_4(diag, shift, allowzeropivot, &zeropivotdetected));
171: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
172: diag += 16;
173: }
174: break;
175: case 5:
176: for (i = 0; i < mbs; i++) {
177: odiag = v + 25 * diag_offset[i];
178: PetscCall(PetscArraycpy(diag, odiag, 25));
179: PetscCall(PetscKernel_A_gets_inverse_A_5(diag, ipvt, work, shift, allowzeropivot, &zeropivotdetected));
180: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
181: diag += 25;
182: }
183: break;
184: case 6:
185: for (i = 0; i < mbs; i++) {
186: odiag = v + 36 * diag_offset[i];
187: PetscCall(PetscArraycpy(diag, odiag, 36));
188: PetscCall(PetscKernel_A_gets_inverse_A_6(diag, shift, allowzeropivot, &zeropivotdetected));
189: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
190: diag += 36;
191: }
192: break;
193: case 7:
194: for (i = 0; i < mbs; i++) {
195: odiag = v + 49 * diag_offset[i];
196: PetscCall(PetscArraycpy(diag, odiag, 49));
197: PetscCall(PetscKernel_A_gets_inverse_A_7(diag, shift, allowzeropivot, &zeropivotdetected));
198: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
199: diag += 49;
200: }
201: break;
202: default:
203: PetscCall(PetscMalloc2(bs, &v_work, bs, &v_pivots));
204: for (i = 0; i < mbs; i++) {
205: odiag = v + bs2 * diag_offset[i];
206: PetscCall(PetscArraycpy(diag, odiag, bs2));
207: PetscCall(PetscKernel_A_gets_inverse_A(bs, diag, v_pivots, v_work, allowzeropivot, &zeropivotdetected));
208: if (zeropivotdetected) A->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT;
209: diag += bs2;
210: }
211: PetscCall(PetscFree2(v_work, v_pivots));
212: }
213: a->idiagvalid = PETSC_TRUE;
214: PetscFunctionReturn(PETSC_SUCCESS);
215: }
217: PetscErrorCode MatSOR_SeqBAIJ(Mat A, Vec bb, PetscReal omega, MatSORType flag, PetscReal fshift, PetscInt its, PetscInt lits, Vec xx)
218: {
219: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
220: PetscScalar *x, *work, *w, *workt, *t;
221: const MatScalar *v, *aa = a->a, *idiag;
222: const PetscScalar *b, *xb;
223: PetscScalar s[7], xw[7] = {0}; /* avoid some compilers thinking xw is uninitialized */
224: PetscInt m = a->mbs, i, i2, nz, bs = A->rmap->bs, bs2 = bs * bs, k, j, idx, it;
225: const PetscInt *diag, *ai = a->i, *aj = a->j, *vi;
227: PetscFunctionBegin;
228: its = its * lits;
229: PetscCheck(!(flag & SOR_EISENSTAT), PETSC_COMM_SELF, PETSC_ERR_SUP, "No support yet for Eisenstat");
230: PetscCheck(its > 0, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "Relaxation requires global its %" PetscInt_FMT " and local its %" PetscInt_FMT " both positive", its, lits);
231: PetscCheck(!fshift, PETSC_COMM_SELF, PETSC_ERR_SUP, "No support for diagonal shift");
232: PetscCheck(omega == 1.0, PETSC_COMM_SELF, PETSC_ERR_SUP, "No support for non-trivial relaxation factor");
233: PetscCheck(!(flag & SOR_APPLY_UPPER) && !(flag & SOR_APPLY_LOWER), PETSC_COMM_SELF, PETSC_ERR_SUP, "No support for applying upper or lower triangular parts");
235: if (!a->idiagvalid) PetscCall(MatInvertBlockDiagonal(A, NULL));
237: if (!m) PetscFunctionReturn(PETSC_SUCCESS);
238: diag = a->diag;
239: idiag = a->idiag;
240: k = PetscMax(A->rmap->n, A->cmap->n);
241: if (!a->mult_work) PetscCall(PetscMalloc1(k + 1, &a->mult_work));
242: if (!a->sor_workt) PetscCall(PetscMalloc1(k, &a->sor_workt));
243: if (!a->sor_work) PetscCall(PetscMalloc1(bs, &a->sor_work));
244: work = a->mult_work;
245: t = a->sor_workt;
246: w = a->sor_work;
248: PetscCall(VecGetArray(xx, &x));
249: PetscCall(VecGetArrayRead(bb, &b));
251: if (flag & SOR_ZERO_INITIAL_GUESS) {
252: if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) {
253: switch (bs) {
254: case 1:
255: PetscKernel_v_gets_A_times_w_1(x, idiag, b);
256: t[0] = b[0];
257: i2 = 1;
258: idiag += 1;
259: for (i = 1; i < m; i++) {
260: v = aa + ai[i];
261: vi = aj + ai[i];
262: nz = diag[i] - ai[i];
263: s[0] = b[i2];
264: for (j = 0; j < nz; j++) {
265: xw[0] = x[vi[j]];
266: PetscKernel_v_gets_v_minus_A_times_w_1(s, (v + j), xw);
267: }
268: t[i2] = s[0];
269: PetscKernel_v_gets_A_times_w_1(xw, idiag, s);
270: x[i2] = xw[0];
271: idiag += 1;
272: i2 += 1;
273: }
274: break;
275: case 2:
276: PetscKernel_v_gets_A_times_w_2(x, idiag, b);
277: t[0] = b[0];
278: t[1] = b[1];
279: i2 = 2;
280: idiag += 4;
281: for (i = 1; i < m; i++) {
282: v = aa + 4 * ai[i];
283: vi = aj + ai[i];
284: nz = diag[i] - ai[i];
285: s[0] = b[i2];
286: s[1] = b[i2 + 1];
287: for (j = 0; j < nz; j++) {
288: idx = 2 * vi[j];
289: it = 4 * j;
290: xw[0] = x[idx];
291: xw[1] = x[1 + idx];
292: PetscKernel_v_gets_v_minus_A_times_w_2(s, (v + it), xw);
293: }
294: t[i2] = s[0];
295: t[i2 + 1] = s[1];
296: PetscKernel_v_gets_A_times_w_2(xw, idiag, s);
297: x[i2] = xw[0];
298: x[i2 + 1] = xw[1];
299: idiag += 4;
300: i2 += 2;
301: }
302: break;
303: case 3:
304: PetscKernel_v_gets_A_times_w_3(x, idiag, b);
305: t[0] = b[0];
306: t[1] = b[1];
307: t[2] = b[2];
308: i2 = 3;
309: idiag += 9;
310: for (i = 1; i < m; i++) {
311: v = aa + 9 * ai[i];
312: vi = aj + ai[i];
313: nz = diag[i] - ai[i];
314: s[0] = b[i2];
315: s[1] = b[i2 + 1];
316: s[2] = b[i2 + 2];
317: while (nz--) {
318: idx = 3 * (*vi++);
319: xw[0] = x[idx];
320: xw[1] = x[1 + idx];
321: xw[2] = x[2 + idx];
322: PetscKernel_v_gets_v_minus_A_times_w_3(s, v, xw);
323: v += 9;
324: }
325: t[i2] = s[0];
326: t[i2 + 1] = s[1];
327: t[i2 + 2] = s[2];
328: PetscKernel_v_gets_A_times_w_3(xw, idiag, s);
329: x[i2] = xw[0];
330: x[i2 + 1] = xw[1];
331: x[i2 + 2] = xw[2];
332: idiag += 9;
333: i2 += 3;
334: }
335: break;
336: case 4:
337: PetscKernel_v_gets_A_times_w_4(x, idiag, b);
338: t[0] = b[0];
339: t[1] = b[1];
340: t[2] = b[2];
341: t[3] = b[3];
342: i2 = 4;
343: idiag += 16;
344: for (i = 1; i < m; i++) {
345: v = aa + 16 * ai[i];
346: vi = aj + ai[i];
347: nz = diag[i] - ai[i];
348: s[0] = b[i2];
349: s[1] = b[i2 + 1];
350: s[2] = b[i2 + 2];
351: s[3] = b[i2 + 3];
352: while (nz--) {
353: idx = 4 * (*vi++);
354: xw[0] = x[idx];
355: xw[1] = x[1 + idx];
356: xw[2] = x[2 + idx];
357: xw[3] = x[3 + idx];
358: PetscKernel_v_gets_v_minus_A_times_w_4(s, v, xw);
359: v += 16;
360: }
361: t[i2] = s[0];
362: t[i2 + 1] = s[1];
363: t[i2 + 2] = s[2];
364: t[i2 + 3] = s[3];
365: PetscKernel_v_gets_A_times_w_4(xw, idiag, s);
366: x[i2] = xw[0];
367: x[i2 + 1] = xw[1];
368: x[i2 + 2] = xw[2];
369: x[i2 + 3] = xw[3];
370: idiag += 16;
371: i2 += 4;
372: }
373: break;
374: case 5:
375: PetscKernel_v_gets_A_times_w_5(x, idiag, b);
376: t[0] = b[0];
377: t[1] = b[1];
378: t[2] = b[2];
379: t[3] = b[3];
380: t[4] = b[4];
381: i2 = 5;
382: idiag += 25;
383: for (i = 1; i < m; i++) {
384: v = aa + 25 * ai[i];
385: vi = aj + ai[i];
386: nz = diag[i] - ai[i];
387: s[0] = b[i2];
388: s[1] = b[i2 + 1];
389: s[2] = b[i2 + 2];
390: s[3] = b[i2 + 3];
391: s[4] = b[i2 + 4];
392: while (nz--) {
393: idx = 5 * (*vi++);
394: xw[0] = x[idx];
395: xw[1] = x[1 + idx];
396: xw[2] = x[2 + idx];
397: xw[3] = x[3 + idx];
398: xw[4] = x[4 + idx];
399: PetscKernel_v_gets_v_minus_A_times_w_5(s, v, xw);
400: v += 25;
401: }
402: t[i2] = s[0];
403: t[i2 + 1] = s[1];
404: t[i2 + 2] = s[2];
405: t[i2 + 3] = s[3];
406: t[i2 + 4] = s[4];
407: PetscKernel_v_gets_A_times_w_5(xw, idiag, s);
408: x[i2] = xw[0];
409: x[i2 + 1] = xw[1];
410: x[i2 + 2] = xw[2];
411: x[i2 + 3] = xw[3];
412: x[i2 + 4] = xw[4];
413: idiag += 25;
414: i2 += 5;
415: }
416: break;
417: case 6:
418: PetscKernel_v_gets_A_times_w_6(x, idiag, b);
419: t[0] = b[0];
420: t[1] = b[1];
421: t[2] = b[2];
422: t[3] = b[3];
423: t[4] = b[4];
424: t[5] = b[5];
425: i2 = 6;
426: idiag += 36;
427: for (i = 1; i < m; i++) {
428: v = aa + 36 * ai[i];
429: vi = aj + ai[i];
430: nz = diag[i] - ai[i];
431: s[0] = b[i2];
432: s[1] = b[i2 + 1];
433: s[2] = b[i2 + 2];
434: s[3] = b[i2 + 3];
435: s[4] = b[i2 + 4];
436: s[5] = b[i2 + 5];
437: while (nz--) {
438: idx = 6 * (*vi++);
439: xw[0] = x[idx];
440: xw[1] = x[1 + idx];
441: xw[2] = x[2 + idx];
442: xw[3] = x[3 + idx];
443: xw[4] = x[4 + idx];
444: xw[5] = x[5 + idx];
445: PetscKernel_v_gets_v_minus_A_times_w_6(s, v, xw);
446: v += 36;
447: }
448: t[i2] = s[0];
449: t[i2 + 1] = s[1];
450: t[i2 + 2] = s[2];
451: t[i2 + 3] = s[3];
452: t[i2 + 4] = s[4];
453: t[i2 + 5] = s[5];
454: PetscKernel_v_gets_A_times_w_6(xw, idiag, s);
455: x[i2] = xw[0];
456: x[i2 + 1] = xw[1];
457: x[i2 + 2] = xw[2];
458: x[i2 + 3] = xw[3];
459: x[i2 + 4] = xw[4];
460: x[i2 + 5] = xw[5];
461: idiag += 36;
462: i2 += 6;
463: }
464: break;
465: case 7:
466: PetscKernel_v_gets_A_times_w_7(x, idiag, b);
467: t[0] = b[0];
468: t[1] = b[1];
469: t[2] = b[2];
470: t[3] = b[3];
471: t[4] = b[4];
472: t[5] = b[5];
473: t[6] = b[6];
474: i2 = 7;
475: idiag += 49;
476: for (i = 1; i < m; i++) {
477: v = aa + 49 * ai[i];
478: vi = aj + ai[i];
479: nz = diag[i] - ai[i];
480: s[0] = b[i2];
481: s[1] = b[i2 + 1];
482: s[2] = b[i2 + 2];
483: s[3] = b[i2 + 3];
484: s[4] = b[i2 + 4];
485: s[5] = b[i2 + 5];
486: s[6] = b[i2 + 6];
487: while (nz--) {
488: idx = 7 * (*vi++);
489: xw[0] = x[idx];
490: xw[1] = x[1 + idx];
491: xw[2] = x[2 + idx];
492: xw[3] = x[3 + idx];
493: xw[4] = x[4 + idx];
494: xw[5] = x[5 + idx];
495: xw[6] = x[6 + idx];
496: PetscKernel_v_gets_v_minus_A_times_w_7(s, v, xw);
497: v += 49;
498: }
499: t[i2] = s[0];
500: t[i2 + 1] = s[1];
501: t[i2 + 2] = s[2];
502: t[i2 + 3] = s[3];
503: t[i2 + 4] = s[4];
504: t[i2 + 5] = s[5];
505: t[i2 + 6] = s[6];
506: PetscKernel_v_gets_A_times_w_7(xw, idiag, s);
507: x[i2] = xw[0];
508: x[i2 + 1] = xw[1];
509: x[i2 + 2] = xw[2];
510: x[i2 + 3] = xw[3];
511: x[i2 + 4] = xw[4];
512: x[i2 + 5] = xw[5];
513: x[i2 + 6] = xw[6];
514: idiag += 49;
515: i2 += 7;
516: }
517: break;
518: default:
519: PetscKernel_w_gets_Ar_times_v(bs, bs, b, idiag, x);
520: PetscCall(PetscArraycpy(t, b, bs));
521: i2 = bs;
522: idiag += bs2;
523: for (i = 1; i < m; i++) {
524: v = aa + bs2 * ai[i];
525: vi = aj + ai[i];
526: nz = diag[i] - ai[i];
528: PetscCall(PetscArraycpy(w, b + i2, bs));
529: /* copy all rows of x that are needed into contiguous space */
530: workt = work;
531: for (j = 0; j < nz; j++) {
532: PetscCall(PetscArraycpy(workt, x + bs * (*vi++), bs));
533: workt += bs;
534: }
535: PetscKernel_w_gets_w_minus_Ar_times_v(bs, bs * nz, w, v, work);
536: PetscCall(PetscArraycpy(t + i2, w, bs));
537: PetscKernel_w_gets_Ar_times_v(bs, bs, w, idiag, x + i2);
539: idiag += bs2;
540: i2 += bs;
541: }
542: break;
543: }
544: /* for logging purposes assume number of nonzero in lower half is 1/2 of total */
545: PetscCall(PetscLogFlops(1.0 * bs2 * a->nz));
546: xb = t;
547: } else xb = b;
548: if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) {
549: idiag = a->idiag + bs2 * (a->mbs - 1);
550: i2 = bs * (m - 1);
551: switch (bs) {
552: case 1:
553: s[0] = xb[i2];
554: PetscKernel_v_gets_A_times_w_1(xw, idiag, s);
555: x[i2] = xw[0];
556: i2 -= 1;
557: for (i = m - 2; i >= 0; i--) {
558: v = aa + (diag[i] + 1);
559: vi = aj + diag[i] + 1;
560: nz = ai[i + 1] - diag[i] - 1;
561: s[0] = xb[i2];
562: for (j = 0; j < nz; j++) {
563: xw[0] = x[vi[j]];
564: PetscKernel_v_gets_v_minus_A_times_w_1(s, (v + j), xw);
565: }
566: PetscKernel_v_gets_A_times_w_1(xw, idiag, s);
567: x[i2] = xw[0];
568: idiag -= 1;
569: i2 -= 1;
570: }
571: break;
572: case 2:
573: s[0] = xb[i2];
574: s[1] = xb[i2 + 1];
575: PetscKernel_v_gets_A_times_w_2(xw, idiag, s);
576: x[i2] = xw[0];
577: x[i2 + 1] = xw[1];
578: i2 -= 2;
579: idiag -= 4;
580: for (i = m - 2; i >= 0; i--) {
581: v = aa + 4 * (diag[i] + 1);
582: vi = aj + diag[i] + 1;
583: nz = ai[i + 1] - diag[i] - 1;
584: s[0] = xb[i2];
585: s[1] = xb[i2 + 1];
586: for (j = 0; j < nz; j++) {
587: idx = 2 * vi[j];
588: it = 4 * j;
589: xw[0] = x[idx];
590: xw[1] = x[1 + idx];
591: PetscKernel_v_gets_v_minus_A_times_w_2(s, (v + it), xw);
592: }
593: PetscKernel_v_gets_A_times_w_2(xw, idiag, s);
594: x[i2] = xw[0];
595: x[i2 + 1] = xw[1];
596: idiag -= 4;
597: i2 -= 2;
598: }
599: break;
600: case 3:
601: s[0] = xb[i2];
602: s[1] = xb[i2 + 1];
603: s[2] = xb[i2 + 2];
604: PetscKernel_v_gets_A_times_w_3(xw, idiag, s);
605: x[i2] = xw[0];
606: x[i2 + 1] = xw[1];
607: x[i2 + 2] = xw[2];
608: i2 -= 3;
609: idiag -= 9;
610: for (i = m - 2; i >= 0; i--) {
611: v = aa + 9 * (diag[i] + 1);
612: vi = aj + diag[i] + 1;
613: nz = ai[i + 1] - diag[i] - 1;
614: s[0] = xb[i2];
615: s[1] = xb[i2 + 1];
616: s[2] = xb[i2 + 2];
617: while (nz--) {
618: idx = 3 * (*vi++);
619: xw[0] = x[idx];
620: xw[1] = x[1 + idx];
621: 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];
627: x[i2 + 1] = xw[1];
628: x[i2 + 2] = xw[2];
629: idiag -= 9;
630: i2 -= 3;
631: }
632: break;
633: case 4:
634: s[0] = xb[i2];
635: s[1] = xb[i2 + 1];
636: s[2] = xb[i2 + 2];
637: s[3] = xb[i2 + 3];
638: PetscKernel_v_gets_A_times_w_4(xw, idiag, s);
639: x[i2] = xw[0];
640: x[i2 + 1] = xw[1];
641: x[i2 + 2] = xw[2];
642: x[i2 + 3] = xw[3];
643: i2 -= 4;
644: idiag -= 16;
645: for (i = m - 2; i >= 0; i--) {
646: v = aa + 16 * (diag[i] + 1);
647: vi = aj + diag[i] + 1;
648: nz = ai[i + 1] - diag[i] - 1;
649: s[0] = xb[i2];
650: s[1] = xb[i2 + 1];
651: s[2] = xb[i2 + 2];
652: s[3] = xb[i2 + 3];
653: while (nz--) {
654: idx = 4 * (*vi++);
655: xw[0] = x[idx];
656: xw[1] = x[1 + idx];
657: xw[2] = x[2 + idx];
658: xw[3] = x[3 + idx];
659: PetscKernel_v_gets_v_minus_A_times_w_4(s, v, xw);
660: v += 16;
661: }
662: PetscKernel_v_gets_A_times_w_4(xw, idiag, s);
663: x[i2] = xw[0];
664: x[i2 + 1] = xw[1];
665: x[i2 + 2] = xw[2];
666: x[i2 + 3] = xw[3];
667: idiag -= 16;
668: i2 -= 4;
669: }
670: break;
671: case 5:
672: s[0] = xb[i2];
673: s[1] = xb[i2 + 1];
674: s[2] = xb[i2 + 2];
675: s[3] = xb[i2 + 3];
676: s[4] = xb[i2 + 4];
677: PetscKernel_v_gets_A_times_w_5(xw, idiag, s);
678: x[i2] = xw[0];
679: x[i2 + 1] = xw[1];
680: x[i2 + 2] = xw[2];
681: x[i2 + 3] = xw[3];
682: x[i2 + 4] = xw[4];
683: i2 -= 5;
684: idiag -= 25;
685: for (i = m - 2; i >= 0; i--) {
686: v = aa + 25 * (diag[i] + 1);
687: vi = aj + diag[i] + 1;
688: nz = ai[i + 1] - diag[i] - 1;
689: s[0] = xb[i2];
690: s[1] = xb[i2 + 1];
691: s[2] = xb[i2 + 2];
692: s[3] = xb[i2 + 3];
693: s[4] = xb[i2 + 4];
694: while (nz--) {
695: idx = 5 * (*vi++);
696: xw[0] = x[idx];
697: xw[1] = x[1 + idx];
698: xw[2] = x[2 + idx];
699: xw[3] = x[3 + idx];
700: xw[4] = x[4 + idx];
701: PetscKernel_v_gets_v_minus_A_times_w_5(s, v, xw);
702: v += 25;
703: }
704: PetscKernel_v_gets_A_times_w_5(xw, idiag, s);
705: x[i2] = xw[0];
706: x[i2 + 1] = xw[1];
707: x[i2 + 2] = xw[2];
708: x[i2 + 3] = xw[3];
709: x[i2 + 4] = xw[4];
710: idiag -= 25;
711: i2 -= 5;
712: }
713: break;
714: case 6:
715: s[0] = xb[i2];
716: s[1] = xb[i2 + 1];
717: s[2] = xb[i2 + 2];
718: s[3] = xb[i2 + 3];
719: s[4] = xb[i2 + 4];
720: s[5] = xb[i2 + 5];
721: PetscKernel_v_gets_A_times_w_6(xw, idiag, s);
722: x[i2] = xw[0];
723: x[i2 + 1] = xw[1];
724: x[i2 + 2] = xw[2];
725: x[i2 + 3] = xw[3];
726: x[i2 + 4] = xw[4];
727: x[i2 + 5] = xw[5];
728: i2 -= 6;
729: idiag -= 36;
730: for (i = m - 2; i >= 0; i--) {
731: v = aa + 36 * (diag[i] + 1);
732: vi = aj + diag[i] + 1;
733: nz = ai[i + 1] - diag[i] - 1;
734: s[0] = xb[i2];
735: s[1] = xb[i2 + 1];
736: s[2] = xb[i2 + 2];
737: s[3] = xb[i2 + 3];
738: s[4] = xb[i2 + 4];
739: s[5] = xb[i2 + 5];
740: while (nz--) {
741: idx = 6 * (*vi++);
742: xw[0] = x[idx];
743: xw[1] = x[1 + idx];
744: xw[2] = x[2 + idx];
745: xw[3] = x[3 + idx];
746: xw[4] = x[4 + idx];
747: xw[5] = x[5 + idx];
748: PetscKernel_v_gets_v_minus_A_times_w_6(s, v, xw);
749: v += 36;
750: }
751: PetscKernel_v_gets_A_times_w_6(xw, idiag, s);
752: x[i2] = xw[0];
753: x[i2 + 1] = xw[1];
754: x[i2 + 2] = xw[2];
755: x[i2 + 3] = xw[3];
756: x[i2 + 4] = xw[4];
757: x[i2 + 5] = xw[5];
758: idiag -= 36;
759: i2 -= 6;
760: }
761: break;
762: case 7:
763: s[0] = xb[i2];
764: s[1] = xb[i2 + 1];
765: s[2] = xb[i2 + 2];
766: s[3] = xb[i2 + 3];
767: s[4] = xb[i2 + 4];
768: s[5] = xb[i2 + 5];
769: s[6] = xb[i2 + 6];
770: PetscKernel_v_gets_A_times_w_7(x, idiag, b);
771: x[i2] = xw[0];
772: x[i2 + 1] = xw[1];
773: x[i2 + 2] = xw[2];
774: x[i2 + 3] = xw[3];
775: x[i2 + 4] = xw[4];
776: x[i2 + 5] = xw[5];
777: x[i2 + 6] = xw[6];
778: i2 -= 7;
779: idiag -= 49;
780: for (i = m - 2; i >= 0; i--) {
781: v = aa + 49 * (diag[i] + 1);
782: vi = aj + diag[i] + 1;
783: nz = ai[i + 1] - diag[i] - 1;
784: s[0] = xb[i2];
785: s[1] = xb[i2 + 1];
786: s[2] = xb[i2 + 2];
787: s[3] = xb[i2 + 3];
788: s[4] = xb[i2 + 4];
789: s[5] = xb[i2 + 5];
790: s[6] = xb[i2 + 6];
791: while (nz--) {
792: idx = 7 * (*vi++);
793: xw[0] = x[idx];
794: xw[1] = x[1 + idx];
795: xw[2] = x[2 + idx];
796: xw[3] = x[3 + idx];
797: xw[4] = x[4 + idx];
798: xw[5] = x[5 + idx];
799: xw[6] = x[6 + idx];
800: PetscKernel_v_gets_v_minus_A_times_w_7(s, v, xw);
801: v += 49;
802: }
803: PetscKernel_v_gets_A_times_w_7(xw, idiag, s);
804: x[i2] = xw[0];
805: x[i2 + 1] = xw[1];
806: x[i2 + 2] = xw[2];
807: x[i2 + 3] = xw[3];
808: x[i2 + 4] = xw[4];
809: x[i2 + 5] = xw[5];
810: x[i2 + 6] = xw[6];
811: idiag -= 49;
812: i2 -= 7;
813: }
814: break;
815: default:
816: PetscCall(PetscArraycpy(w, xb + i2, bs));
817: PetscKernel_w_gets_Ar_times_v(bs, bs, w, idiag, x + i2);
818: i2 -= bs;
819: idiag -= bs2;
820: for (i = m - 2; i >= 0; i--) {
821: v = aa + bs2 * (diag[i] + 1);
822: vi = aj + diag[i] + 1;
823: nz = ai[i + 1] - diag[i] - 1;
825: PetscCall(PetscArraycpy(w, xb + i2, bs));
826: /* copy all rows of x that are needed into contiguous space */
827: workt = work;
828: for (j = 0; j < nz; j++) {
829: PetscCall(PetscArraycpy(workt, x + bs * (*vi++), bs));
830: workt += bs;
831: }
832: PetscKernel_w_gets_w_minus_Ar_times_v(bs, bs * nz, w, v, work);
833: PetscKernel_w_gets_Ar_times_v(bs, bs, w, idiag, x + i2);
835: idiag -= bs2;
836: i2 -= bs;
837: }
838: break;
839: }
840: PetscCall(PetscLogFlops(1.0 * bs2 * (a->nz)));
841: }
842: its--;
843: }
844: while (its--) {
845: if (flag & SOR_FORWARD_SWEEP || flag & SOR_LOCAL_FORWARD_SWEEP) {
846: idiag = a->idiag;
847: i2 = 0;
848: switch (bs) {
849: case 1:
850: for (i = 0; i < m; i++) {
851: v = aa + ai[i];
852: vi = aj + ai[i];
853: nz = ai[i + 1] - ai[i];
854: s[0] = b[i2];
855: for (j = 0; j < nz; j++) {
856: xw[0] = x[vi[j]];
857: PetscKernel_v_gets_v_minus_A_times_w_1(s, (v + j), xw);
858: }
859: PetscKernel_v_gets_A_times_w_1(xw, idiag, s);
860: x[i2] += xw[0];
861: idiag += 1;
862: i2 += 1;
863: }
864: break;
865: case 2:
866: for (i = 0; i < m; i++) {
867: v = aa + 4 * ai[i];
868: vi = aj + ai[i];
869: nz = ai[i + 1] - ai[i];
870: s[0] = b[i2];
871: s[1] = b[i2 + 1];
872: for (j = 0; j < nz; j++) {
873: idx = 2 * vi[j];
874: it = 4 * j;
875: xw[0] = x[idx];
876: xw[1] = x[1 + idx];
877: PetscKernel_v_gets_v_minus_A_times_w_2(s, (v + it), xw);
878: }
879: PetscKernel_v_gets_A_times_w_2(xw, idiag, s);
880: x[i2] += xw[0];
881: x[i2 + 1] += xw[1];
882: idiag += 4;
883: i2 += 2;
884: }
885: break;
886: case 3:
887: for (i = 0; i < m; i++) {
888: v = aa + 9 * ai[i];
889: vi = aj + ai[i];
890: nz = ai[i + 1] - ai[i];
891: s[0] = b[i2];
892: s[1] = b[i2 + 1];
893: s[2] = b[i2 + 2];
894: while (nz--) {
895: idx = 3 * (*vi++);
896: xw[0] = x[idx];
897: xw[1] = x[1 + idx];
898: xw[2] = x[2 + idx];
899: PetscKernel_v_gets_v_minus_A_times_w_3(s, v, xw);
900: v += 9;
901: }
902: PetscKernel_v_gets_A_times_w_3(xw, idiag, s);
903: x[i2] += xw[0];
904: x[i2 + 1] += xw[1];
905: x[i2 + 2] += xw[2];
906: idiag += 9;
907: i2 += 3;
908: }
909: break;
910: case 4:
911: for (i = 0; i < m; i++) {
912: v = aa + 16 * ai[i];
913: vi = aj + ai[i];
914: nz = ai[i + 1] - ai[i];
915: s[0] = b[i2];
916: s[1] = b[i2 + 1];
917: s[2] = b[i2 + 2];
918: s[3] = b[i2 + 3];
919: while (nz--) {
920: idx = 4 * (*vi++);
921: xw[0] = x[idx];
922: xw[1] = x[1 + idx];
923: xw[2] = x[2 + idx];
924: xw[3] = x[3 + idx];
925: PetscKernel_v_gets_v_minus_A_times_w_4(s, v, xw);
926: v += 16;
927: }
928: PetscKernel_v_gets_A_times_w_4(xw, idiag, s);
929: x[i2] += xw[0];
930: x[i2 + 1] += xw[1];
931: x[i2 + 2] += xw[2];
932: x[i2 + 3] += xw[3];
933: idiag += 16;
934: i2 += 4;
935: }
936: break;
937: case 5:
938: for (i = 0; i < m; i++) {
939: v = aa + 25 * ai[i];
940: vi = aj + ai[i];
941: nz = ai[i + 1] - ai[i];
942: s[0] = b[i2];
943: s[1] = b[i2 + 1];
944: s[2] = b[i2 + 2];
945: s[3] = b[i2 + 3];
946: s[4] = b[i2 + 4];
947: while (nz--) {
948: idx = 5 * (*vi++);
949: xw[0] = x[idx];
950: xw[1] = x[1 + idx];
951: xw[2] = x[2 + idx];
952: xw[3] = x[3 + idx];
953: xw[4] = x[4 + idx];
954: PetscKernel_v_gets_v_minus_A_times_w_5(s, v, xw);
955: v += 25;
956: }
957: PetscKernel_v_gets_A_times_w_5(xw, idiag, s);
958: x[i2] += xw[0];
959: x[i2 + 1] += xw[1];
960: x[i2 + 2] += xw[2];
961: x[i2 + 3] += xw[3];
962: x[i2 + 4] += xw[4];
963: idiag += 25;
964: i2 += 5;
965: }
966: break;
967: case 6:
968: for (i = 0; i < m; i++) {
969: v = aa + 36 * ai[i];
970: vi = aj + ai[i];
971: nz = ai[i + 1] - ai[i];
972: s[0] = b[i2];
973: s[1] = b[i2 + 1];
974: s[2] = b[i2 + 2];
975: s[3] = b[i2 + 3];
976: s[4] = b[i2 + 4];
977: s[5] = b[i2 + 5];
978: while (nz--) {
979: idx = 6 * (*vi++);
980: xw[0] = x[idx];
981: xw[1] = x[1 + idx];
982: xw[2] = x[2 + idx];
983: xw[3] = x[3 + idx];
984: xw[4] = x[4 + idx];
985: xw[5] = x[5 + idx];
986: PetscKernel_v_gets_v_minus_A_times_w_6(s, v, xw);
987: v += 36;
988: }
989: PetscKernel_v_gets_A_times_w_6(xw, idiag, s);
990: x[i2] += xw[0];
991: x[i2 + 1] += xw[1];
992: x[i2 + 2] += xw[2];
993: x[i2 + 3] += xw[3];
994: x[i2 + 4] += xw[4];
995: x[i2 + 5] += xw[5];
996: idiag += 36;
997: i2 += 6;
998: }
999: break;
1000: case 7:
1001: for (i = 0; i < m; i++) {
1002: v = aa + 49 * ai[i];
1003: vi = aj + ai[i];
1004: nz = ai[i + 1] - ai[i];
1005: s[0] = b[i2];
1006: s[1] = b[i2 + 1];
1007: s[2] = b[i2 + 2];
1008: s[3] = b[i2 + 3];
1009: s[4] = b[i2 + 4];
1010: s[5] = b[i2 + 5];
1011: s[6] = b[i2 + 6];
1012: while (nz--) {
1013: idx = 7 * (*vi++);
1014: xw[0] = x[idx];
1015: xw[1] = x[1 + idx];
1016: xw[2] = x[2 + idx];
1017: xw[3] = x[3 + idx];
1018: xw[4] = x[4 + idx];
1019: xw[5] = x[5 + idx];
1020: xw[6] = x[6 + idx];
1021: PetscKernel_v_gets_v_minus_A_times_w_7(s, v, xw);
1022: v += 49;
1023: }
1024: PetscKernel_v_gets_A_times_w_7(xw, idiag, s);
1025: x[i2] += xw[0];
1026: x[i2 + 1] += xw[1];
1027: x[i2 + 2] += xw[2];
1028: x[i2 + 3] += xw[3];
1029: x[i2 + 4] += xw[4];
1030: x[i2 + 5] += xw[5];
1031: x[i2 + 6] += xw[6];
1032: idiag += 49;
1033: i2 += 7;
1034: }
1035: break;
1036: default:
1037: for (i = 0; i < m; i++) {
1038: v = aa + bs2 * ai[i];
1039: vi = aj + ai[i];
1040: nz = ai[i + 1] - ai[i];
1042: PetscCall(PetscArraycpy(w, b + i2, bs));
1043: /* copy all rows of x that are needed into contiguous space */
1044: workt = work;
1045: for (j = 0; j < nz; j++) {
1046: PetscCall(PetscArraycpy(workt, x + bs * (*vi++), bs));
1047: workt += bs;
1048: }
1049: PetscKernel_w_gets_w_minus_Ar_times_v(bs, bs * nz, w, v, work);
1050: PetscKernel_w_gets_w_plus_Ar_times_v(bs, bs, w, idiag, x + i2);
1052: idiag += bs2;
1053: i2 += bs;
1054: }
1055: break;
1056: }
1057: PetscCall(PetscLogFlops(2.0 * bs2 * a->nz));
1058: }
1059: if (flag & SOR_BACKWARD_SWEEP || flag & SOR_LOCAL_BACKWARD_SWEEP) {
1060: idiag = a->idiag + bs2 * (a->mbs - 1);
1061: i2 = bs * (m - 1);
1062: switch (bs) {
1063: case 1:
1064: for (i = m - 1; i >= 0; i--) {
1065: v = aa + ai[i];
1066: vi = aj + ai[i];
1067: nz = ai[i + 1] - ai[i];
1068: s[0] = b[i2];
1069: for (j = 0; j < nz; j++) {
1070: xw[0] = x[vi[j]];
1071: PetscKernel_v_gets_v_minus_A_times_w_1(s, (v + j), xw);
1072: }
1073: PetscKernel_v_gets_A_times_w_1(xw, idiag, s);
1074: x[i2] += xw[0];
1075: idiag -= 1;
1076: i2 -= 1;
1077: }
1078: break;
1079: case 2:
1080: for (i = m - 1; i >= 0; i--) {
1081: v = aa + 4 * ai[i];
1082: vi = aj + ai[i];
1083: nz = ai[i + 1] - ai[i];
1084: s[0] = b[i2];
1085: s[1] = b[i2 + 1];
1086: for (j = 0; j < nz; j++) {
1087: idx = 2 * vi[j];
1088: it = 4 * j;
1089: xw[0] = x[idx];
1090: xw[1] = x[1 + idx];
1091: PetscKernel_v_gets_v_minus_A_times_w_2(s, (v + it), xw);
1092: }
1093: PetscKernel_v_gets_A_times_w_2(xw, idiag, s);
1094: x[i2] += xw[0];
1095: x[i2 + 1] += xw[1];
1096: idiag -= 4;
1097: i2 -= 2;
1098: }
1099: break;
1100: case 3:
1101: for (i = m - 1; i >= 0; i--) {
1102: v = aa + 9 * ai[i];
1103: vi = aj + ai[i];
1104: nz = ai[i + 1] - ai[i];
1105: s[0] = b[i2];
1106: s[1] = b[i2 + 1];
1107: s[2] = b[i2 + 2];
1108: while (nz--) {
1109: idx = 3 * (*vi++);
1110: xw[0] = x[idx];
1111: xw[1] = x[1 + idx];
1112: xw[2] = x[2 + idx];
1113: PetscKernel_v_gets_v_minus_A_times_w_3(s, v, xw);
1114: v += 9;
1115: }
1116: PetscKernel_v_gets_A_times_w_3(xw, idiag, s);
1117: x[i2] += xw[0];
1118: x[i2 + 1] += xw[1];
1119: x[i2 + 2] += xw[2];
1120: idiag -= 9;
1121: i2 -= 3;
1122: }
1123: break;
1124: case 4:
1125: for (i = m - 1; i >= 0; i--) {
1126: v = aa + 16 * ai[i];
1127: vi = aj + ai[i];
1128: nz = ai[i + 1] - ai[i];
1129: s[0] = b[i2];
1130: s[1] = b[i2 + 1];
1131: s[2] = b[i2 + 2];
1132: s[3] = b[i2 + 3];
1133: while (nz--) {
1134: idx = 4 * (*vi++);
1135: xw[0] = x[idx];
1136: xw[1] = x[1 + idx];
1137: xw[2] = x[2 + idx];
1138: xw[3] = x[3 + idx];
1139: PetscKernel_v_gets_v_minus_A_times_w_4(s, v, xw);
1140: v += 16;
1141: }
1142: PetscKernel_v_gets_A_times_w_4(xw, idiag, s);
1143: x[i2] += xw[0];
1144: x[i2 + 1] += xw[1];
1145: x[i2 + 2] += xw[2];
1146: x[i2 + 3] += xw[3];
1147: idiag -= 16;
1148: i2 -= 4;
1149: }
1150: break;
1151: case 5:
1152: for (i = m - 1; i >= 0; i--) {
1153: v = aa + 25 * ai[i];
1154: vi = aj + ai[i];
1155: nz = ai[i + 1] - ai[i];
1156: s[0] = b[i2];
1157: s[1] = b[i2 + 1];
1158: s[2] = b[i2 + 2];
1159: s[3] = b[i2 + 3];
1160: s[4] = b[i2 + 4];
1161: while (nz--) {
1162: idx = 5 * (*vi++);
1163: xw[0] = x[idx];
1164: xw[1] = x[1 + idx];
1165: xw[2] = x[2 + idx];
1166: xw[3] = x[3 + idx];
1167: xw[4] = x[4 + idx];
1168: PetscKernel_v_gets_v_minus_A_times_w_5(s, v, xw);
1169: v += 25;
1170: }
1171: PetscKernel_v_gets_A_times_w_5(xw, idiag, s);
1172: x[i2] += xw[0];
1173: x[i2 + 1] += xw[1];
1174: x[i2 + 2] += xw[2];
1175: x[i2 + 3] += xw[3];
1176: x[i2 + 4] += xw[4];
1177: idiag -= 25;
1178: i2 -= 5;
1179: }
1180: break;
1181: case 6:
1182: for (i = m - 1; i >= 0; i--) {
1183: v = aa + 36 * ai[i];
1184: vi = aj + ai[i];
1185: nz = ai[i + 1] - ai[i];
1186: s[0] = b[i2];
1187: s[1] = b[i2 + 1];
1188: s[2] = b[i2 + 2];
1189: s[3] = b[i2 + 3];
1190: s[4] = b[i2 + 4];
1191: s[5] = b[i2 + 5];
1192: while (nz--) {
1193: idx = 6 * (*vi++);
1194: xw[0] = x[idx];
1195: xw[1] = x[1 + idx];
1196: xw[2] = x[2 + idx];
1197: xw[3] = x[3 + idx];
1198: xw[4] = x[4 + idx];
1199: xw[5] = x[5 + idx];
1200: PetscKernel_v_gets_v_minus_A_times_w_6(s, v, xw);
1201: v += 36;
1202: }
1203: PetscKernel_v_gets_A_times_w_6(xw, idiag, s);
1204: x[i2] += xw[0];
1205: x[i2 + 1] += xw[1];
1206: x[i2 + 2] += xw[2];
1207: x[i2 + 3] += xw[3];
1208: x[i2 + 4] += xw[4];
1209: x[i2 + 5] += xw[5];
1210: idiag -= 36;
1211: i2 -= 6;
1212: }
1213: break;
1214: case 7:
1215: for (i = m - 1; i >= 0; i--) {
1216: v = aa + 49 * ai[i];
1217: vi = aj + ai[i];
1218: nz = ai[i + 1] - ai[i];
1219: s[0] = b[i2];
1220: s[1] = b[i2 + 1];
1221: s[2] = b[i2 + 2];
1222: s[3] = b[i2 + 3];
1223: s[4] = b[i2 + 4];
1224: s[5] = b[i2 + 5];
1225: s[6] = b[i2 + 6];
1226: while (nz--) {
1227: idx = 7 * (*vi++);
1228: xw[0] = x[idx];
1229: xw[1] = x[1 + idx];
1230: xw[2] = x[2 + idx];
1231: xw[3] = x[3 + idx];
1232: xw[4] = x[4 + idx];
1233: xw[5] = x[5 + idx];
1234: xw[6] = x[6 + idx];
1235: PetscKernel_v_gets_v_minus_A_times_w_7(s, v, xw);
1236: v += 49;
1237: }
1238: PetscKernel_v_gets_A_times_w_7(xw, idiag, s);
1239: x[i2] += xw[0];
1240: x[i2 + 1] += xw[1];
1241: x[i2 + 2] += xw[2];
1242: x[i2 + 3] += xw[3];
1243: x[i2 + 4] += xw[4];
1244: x[i2 + 5] += xw[5];
1245: x[i2 + 6] += xw[6];
1246: idiag -= 49;
1247: i2 -= 7;
1248: }
1249: break;
1250: default:
1251: for (i = m - 1; i >= 0; i--) {
1252: v = aa + bs2 * ai[i];
1253: vi = aj + ai[i];
1254: nz = ai[i + 1] - ai[i];
1256: PetscCall(PetscArraycpy(w, b + i2, bs));
1257: /* copy all rows of x that are needed into contiguous space */
1258: workt = work;
1259: for (j = 0; j < nz; j++) {
1260: PetscCall(PetscArraycpy(workt, x + bs * (*vi++), bs));
1261: workt += bs;
1262: }
1263: PetscKernel_w_gets_w_minus_Ar_times_v(bs, bs * nz, w, v, work);
1264: PetscKernel_w_gets_w_plus_Ar_times_v(bs, bs, w, idiag, x + i2);
1266: idiag -= bs2;
1267: i2 -= bs;
1268: }
1269: break;
1270: }
1271: PetscCall(PetscLogFlops(2.0 * bs2 * (a->nz)));
1272: }
1273: }
1274: PetscCall(VecRestoreArray(xx, &x));
1275: PetscCall(VecRestoreArrayRead(bb, &b));
1276: PetscFunctionReturn(PETSC_SUCCESS);
1277: }
1279: /*
1280: Special version for direct calls from Fortran (Used in PETSc-fun3d)
1281: */
1282: #if defined(PETSC_HAVE_FORTRAN_CAPS)
1283: #define matsetvaluesblocked4_ MATSETVALUESBLOCKED4
1284: #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE)
1285: #define matsetvaluesblocked4_ matsetvaluesblocked4
1286: #endif
1288: PETSC_EXTERN void matsetvaluesblocked4_(Mat *AA, PetscInt *mm, const PetscInt im[], PetscInt *nn, const PetscInt in[], const PetscScalar v[])
1289: {
1290: Mat A = *AA;
1291: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
1292: PetscInt *rp, k, low, high, t, ii, jj, row, nrow, i, col, l, N, m = *mm, n = *nn;
1293: PetscInt *ai = a->i, *ailen = a->ilen;
1294: PetscInt *aj = a->j, stepval, lastcol = -1;
1295: const PetscScalar *value = v;
1296: MatScalar *ap, *aa = a->a, *bap;
1298: PetscFunctionBegin;
1299: if (A->rmap->bs != 4) SETERRABORT(PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONG, "Can only be called with a block size of 4");
1300: stepval = (n - 1) * 4;
1301: for (k = 0; k < m; k++) { /* loop over added rows */
1302: row = im[k];
1303: rp = aj + ai[row];
1304: ap = aa + 16 * ai[row];
1305: nrow = ailen[row];
1306: low = 0;
1307: high = nrow;
1308: for (l = 0; l < n; l++) { /* loop over added columns */
1309: col = in[l];
1310: if (col <= lastcol) low = 0;
1311: else high = nrow;
1312: lastcol = col;
1313: value = v + k * (stepval + 4 + l) * 4;
1314: while (high - low > 7) {
1315: t = (low + high) / 2;
1316: if (rp[t] > col) high = t;
1317: else low = t;
1318: }
1319: for (i = low; i < high; i++) {
1320: if (rp[i] > col) break;
1321: if (rp[i] == col) {
1322: bap = ap + 16 * i;
1323: for (ii = 0; ii < 4; ii++, value += stepval) {
1324: for (jj = ii; jj < 16; jj += 4) bap[jj] += *value++;
1325: }
1326: goto noinsert2;
1327: }
1328: }
1329: N = nrow++ - 1;
1330: high++; /* added new column index thus must search to one higher than before */
1331: /* shift up all the later entries in this row */
1332: for (ii = N; ii >= i; ii--) {
1333: rp[ii + 1] = rp[ii];
1334: PetscCallVoid(PetscArraycpy(ap + 16 * (ii + 1), ap + 16 * (ii), 16));
1335: }
1336: if (N >= i) PetscCallVoid(PetscArrayzero(ap + 16 * i, 16));
1337: rp[i] = col;
1338: bap = ap + 16 * i;
1339: for (ii = 0; ii < 4; ii++, value += stepval) {
1340: for (jj = ii; jj < 16; jj += 4) bap[jj] = *value++;
1341: }
1342: noinsert2:;
1343: low = i;
1344: }
1345: ailen[row] = nrow;
1346: }
1347: PetscFunctionReturnVoid();
1348: }
1350: #if defined(PETSC_HAVE_FORTRAN_CAPS)
1351: #define matsetvalues4_ MATSETVALUES4
1352: #elif !defined(PETSC_HAVE_FORTRAN_UNDERSCORE)
1353: #define matsetvalues4_ matsetvalues4
1354: #endif
1356: PETSC_EXTERN void matsetvalues4_(Mat *AA, PetscInt *mm, PetscInt *im, PetscInt *nn, PetscInt *in, PetscScalar *v)
1357: {
1358: Mat A = *AA;
1359: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
1360: PetscInt *rp, k, low, high, t, row, nrow, i, col, l, N, n = *nn, m = *mm;
1361: PetscInt *ai = a->i, *ailen = a->ilen;
1362: PetscInt *aj = a->j, brow, bcol;
1363: PetscInt ridx, cidx, lastcol = -1;
1364: MatScalar *ap, value, *aa = a->a, *bap;
1366: PetscFunctionBegin;
1367: for (k = 0; k < m; k++) { /* loop over added rows */
1368: row = im[k];
1369: brow = row / 4;
1370: rp = aj + ai[brow];
1371: ap = aa + 16 * ai[brow];
1372: nrow = ailen[brow];
1373: low = 0;
1374: high = nrow;
1375: for (l = 0; l < n; l++) { /* loop over added columns */
1376: col = in[l];
1377: bcol = col / 4;
1378: ridx = row % 4;
1379: cidx = col % 4;
1380: value = v[l + k * n];
1381: if (col <= lastcol) low = 0;
1382: else high = nrow;
1383: lastcol = col;
1384: while (high - low > 7) {
1385: t = (low + high) / 2;
1386: if (rp[t] > bcol) high = t;
1387: else low = t;
1388: }
1389: for (i = low; i < high; i++) {
1390: if (rp[i] > bcol) break;
1391: if (rp[i] == bcol) {
1392: bap = ap + 16 * i + 4 * cidx + ridx;
1393: *bap += value;
1394: goto noinsert1;
1395: }
1396: }
1397: N = nrow++ - 1;
1398: high++; /* added new column thus must search to one higher than before */
1399: /* shift up all the later entries in this row */
1400: PetscCallVoid(PetscArraymove(rp + i + 1, rp + i, N - i + 1));
1401: PetscCallVoid(PetscArraymove(ap + 16 * i + 16, ap + 16 * i, 16 * (N - i + 1)));
1402: PetscCallVoid(PetscArrayzero(ap + 16 * i, 16));
1403: rp[i] = bcol;
1404: ap[16 * i + 4 * cidx + ridx] = value;
1405: noinsert1:;
1406: low = i;
1407: }
1408: ailen[brow] = nrow;
1409: }
1410: PetscFunctionReturnVoid();
1411: }
1413: /*
1414: Checks for missing diagonals
1415: */
1416: PetscErrorCode MatMissingDiagonal_SeqBAIJ(Mat A, PetscBool *missing, PetscInt *d)
1417: {
1418: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
1419: PetscInt *diag, *ii = a->i, i;
1421: PetscFunctionBegin;
1422: PetscCall(MatMarkDiagonal_SeqBAIJ(A));
1423: *missing = PETSC_FALSE;
1424: if (A->rmap->n > 0 && !ii) {
1425: *missing = PETSC_TRUE;
1426: if (d) *d = 0;
1427: PetscCall(PetscInfo(A, "Matrix has no entries therefore is missing diagonal\n"));
1428: } else {
1429: PetscInt n;
1430: n = PetscMin(a->mbs, a->nbs);
1431: diag = a->diag;
1432: for (i = 0; i < n; i++) {
1433: if (diag[i] >= ii[i + 1]) {
1434: *missing = PETSC_TRUE;
1435: if (d) *d = i;
1436: PetscCall(PetscInfo(A, "Matrix is missing block diagonal number %" PetscInt_FMT "\n", i));
1437: break;
1438: }
1439: }
1440: }
1441: PetscFunctionReturn(PETSC_SUCCESS);
1442: }
1444: PetscErrorCode MatMarkDiagonal_SeqBAIJ(Mat A)
1445: {
1446: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
1447: PetscInt i, j, m = a->mbs;
1449: PetscFunctionBegin;
1450: if (!a->diag) {
1451: PetscCall(PetscMalloc1(m, &a->diag));
1452: a->free_diag = PETSC_TRUE;
1453: }
1454: for (i = 0; i < m; i++) {
1455: a->diag[i] = a->i[i + 1];
1456: for (j = a->i[i]; j < a->i[i + 1]; j++) {
1457: if (a->j[j] == i) {
1458: a->diag[i] = j;
1459: break;
1460: }
1461: }
1462: }
1463: PetscFunctionReturn(PETSC_SUCCESS);
1464: }
1466: static PetscErrorCode MatGetRowIJ_SeqBAIJ(Mat A, PetscInt oshift, PetscBool symmetric, PetscBool blockcompressed, PetscInt *nn, const PetscInt *inia[], const PetscInt *inja[], PetscBool *done)
1467: {
1468: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
1469: PetscInt i, j, n = a->mbs, nz = a->i[n], *tia, *tja, bs = A->rmap->bs, k, l, cnt;
1470: PetscInt **ia = (PetscInt **)inia, **ja = (PetscInt **)inja;
1472: PetscFunctionBegin;
1473: *nn = n;
1474: if (!ia) PetscFunctionReturn(PETSC_SUCCESS);
1475: if (symmetric) {
1476: PetscCall(MatToSymmetricIJ_SeqAIJ(n, a->i, a->j, PETSC_TRUE, 0, 0, &tia, &tja));
1477: nz = tia[n];
1478: } else {
1479: tia = a->i;
1480: tja = a->j;
1481: }
1483: if (!blockcompressed && bs > 1) {
1484: (*nn) *= bs;
1485: /* malloc & create the natural set of indices */
1486: PetscCall(PetscMalloc1((n + 1) * bs, ia));
1487: if (n) {
1488: (*ia)[0] = oshift;
1489: for (j = 1; j < bs; j++) (*ia)[j] = (tia[1] - tia[0]) * bs + (*ia)[j - 1];
1490: }
1492: for (i = 1; i < n; i++) {
1493: (*ia)[i * bs] = (tia[i] - tia[i - 1]) * bs + (*ia)[i * bs - 1];
1494: for (j = 1; j < bs; j++) (*ia)[i * bs + j] = (tia[i + 1] - tia[i]) * bs + (*ia)[i * bs + j - 1];
1495: }
1496: if (n) (*ia)[n * bs] = (tia[n] - tia[n - 1]) * bs + (*ia)[n * bs - 1];
1498: if (inja) {
1499: PetscCall(PetscMalloc1(nz * bs * bs, ja));
1500: cnt = 0;
1501: for (i = 0; i < n; i++) {
1502: for (j = 0; j < bs; j++) {
1503: for (k = tia[i]; k < tia[i + 1]; k++) {
1504: for (l = 0; l < bs; l++) (*ja)[cnt++] = bs * tja[k] + l;
1505: }
1506: }
1507: }
1508: }
1510: if (symmetric) { /* deallocate memory allocated in MatToSymmetricIJ_SeqAIJ() */
1511: PetscCall(PetscFree(tia));
1512: PetscCall(PetscFree(tja));
1513: }
1514: } else if (oshift == 1) {
1515: if (symmetric) {
1516: nz = tia[A->rmap->n / bs];
1517: /* add 1 to i and j indices */
1518: for (i = 0; i < A->rmap->n / bs + 1; i++) tia[i] = tia[i] + 1;
1519: *ia = tia;
1520: if (ja) {
1521: for (i = 0; i < nz; i++) tja[i] = tja[i] + 1;
1522: *ja = tja;
1523: }
1524: } else {
1525: nz = a->i[A->rmap->n / bs];
1526: /* malloc space and add 1 to i and j indices */
1527: PetscCall(PetscMalloc1(A->rmap->n / bs + 1, ia));
1528: for (i = 0; i < A->rmap->n / bs + 1; i++) (*ia)[i] = a->i[i] + 1;
1529: if (ja) {
1530: PetscCall(PetscMalloc1(nz, ja));
1531: for (i = 0; i < nz; i++) (*ja)[i] = a->j[i] + 1;
1532: }
1533: }
1534: } else {
1535: *ia = tia;
1536: if (ja) *ja = tja;
1537: }
1538: PetscFunctionReturn(PETSC_SUCCESS);
1539: }
1541: static PetscErrorCode MatRestoreRowIJ_SeqBAIJ(Mat A, PetscInt oshift, PetscBool symmetric, PetscBool blockcompressed, PetscInt *nn, const PetscInt *ia[], const PetscInt *ja[], PetscBool *done)
1542: {
1543: PetscFunctionBegin;
1544: if (!ia) PetscFunctionReturn(PETSC_SUCCESS);
1545: if ((!blockcompressed && A->rmap->bs > 1) || (symmetric || oshift == 1)) {
1546: PetscCall(PetscFree(*ia));
1547: if (ja) PetscCall(PetscFree(*ja));
1548: }
1549: PetscFunctionReturn(PETSC_SUCCESS);
1550: }
1552: PetscErrorCode MatDestroy_SeqBAIJ(Mat A)
1553: {
1554: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
1556: PetscFunctionBegin;
1557: if (A->hash_active) {
1558: PetscInt bs;
1559: PetscCall(PetscMemcpy(&A->ops, &a->cops, sizeof(*(A->ops))));
1560: PetscCall(PetscHMapIJVDestroy(&a->ht));
1561: PetscCall(MatGetBlockSize(A, &bs));
1562: if (bs > 1) PetscCall(PetscHSetIJDestroy(&a->bht));
1563: PetscCall(PetscFree(a->dnz));
1564: PetscCall(PetscFree(a->bdnz));
1565: A->hash_active = PETSC_FALSE;
1566: }
1567: #if defined(PETSC_USE_LOG)
1568: PetscCall(PetscLogObjectState((PetscObject)A, "Rows=%" PetscInt_FMT ", Cols=%" PetscInt_FMT ", NZ=%" PetscInt_FMT, A->rmap->N, A->cmap->n, a->nz));
1569: #endif
1570: PetscCall(MatSeqXAIJFreeAIJ(A, &a->a, &a->j, &a->i));
1571: PetscCall(ISDestroy(&a->row));
1572: PetscCall(ISDestroy(&a->col));
1573: if (a->free_diag) PetscCall(PetscFree(a->diag));
1574: PetscCall(PetscFree(a->idiag));
1575: if (a->free_imax_ilen) PetscCall(PetscFree2(a->imax, a->ilen));
1576: PetscCall(PetscFree(a->solve_work));
1577: PetscCall(PetscFree(a->mult_work));
1578: PetscCall(PetscFree(a->sor_workt));
1579: PetscCall(PetscFree(a->sor_work));
1580: PetscCall(ISDestroy(&a->icol));
1581: PetscCall(PetscFree(a->saved_values));
1582: PetscCall(PetscFree2(a->compressedrow.i, a->compressedrow.rindex));
1584: PetscCall(MatDestroy(&a->sbaijMat));
1585: PetscCall(MatDestroy(&a->parent));
1586: PetscCall(PetscFree(A->data));
1588: PetscCall(PetscObjectChangeTypeName((PetscObject)A, NULL));
1589: PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatSeqBAIJGetArray_C", NULL));
1590: PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatSeqBAIJRestoreArray_C", NULL));
1591: PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatStoreValues_C", NULL));
1592: PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatRetrieveValues_C", NULL));
1593: PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatSeqBAIJSetColumnIndices_C", NULL));
1594: PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatConvert_seqbaij_seqaij_C", NULL));
1595: PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatConvert_seqbaij_seqsbaij_C", NULL));
1596: PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatSeqBAIJSetPreallocation_C", NULL));
1597: PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatSeqBAIJSetPreallocationCSR_C", NULL));
1598: PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatConvert_seqbaij_seqbstrm_C", NULL));
1599: PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatIsTranspose_C", NULL));
1600: #if defined(PETSC_HAVE_HYPRE)
1601: PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatConvert_seqbaij_hypre_C", NULL));
1602: #endif
1603: PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatConvert_seqbaij_is_C", NULL));
1604: PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatFactorGetSolverType_C", NULL));
1605: PetscFunctionReturn(PETSC_SUCCESS);
1606: }
1608: PetscErrorCode MatSetOption_SeqBAIJ(Mat A, MatOption op, PetscBool flg)
1609: {
1610: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
1612: PetscFunctionBegin;
1613: switch (op) {
1614: case MAT_ROW_ORIENTED:
1615: a->roworiented = flg;
1616: break;
1617: case MAT_KEEP_NONZERO_PATTERN:
1618: a->keepnonzeropattern = flg;
1619: break;
1620: case MAT_NEW_NONZERO_LOCATIONS:
1621: a->nonew = (flg ? 0 : 1);
1622: break;
1623: case MAT_NEW_NONZERO_LOCATION_ERR:
1624: a->nonew = (flg ? -1 : 0);
1625: break;
1626: case MAT_NEW_NONZERO_ALLOCATION_ERR:
1627: a->nonew = (flg ? -2 : 0);
1628: break;
1629: case MAT_UNUSED_NONZERO_LOCATION_ERR:
1630: a->nounused = (flg ? -1 : 0);
1631: break;
1632: case MAT_FORCE_DIAGONAL_ENTRIES:
1633: case MAT_IGNORE_OFF_PROC_ENTRIES:
1634: case MAT_USE_HASH_TABLE:
1635: case MAT_SORTED_FULL:
1636: PetscCall(PetscInfo(A, "Option %s ignored\n", MatOptions[op]));
1637: break;
1638: case MAT_SPD:
1639: case MAT_SYMMETRIC:
1640: case MAT_STRUCTURALLY_SYMMETRIC:
1641: case MAT_HERMITIAN:
1642: case MAT_SYMMETRY_ETERNAL:
1643: case MAT_STRUCTURAL_SYMMETRY_ETERNAL:
1644: case MAT_SUBMAT_SINGLEIS:
1645: case MAT_STRUCTURE_ONLY:
1646: case MAT_SPD_ETERNAL:
1647: /* if the diagonal matrix is square it inherits some of the properties above */
1648: break;
1649: default:
1650: SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP, "unknown option %d", op);
1651: }
1652: PetscFunctionReturn(PETSC_SUCCESS);
1653: }
1655: /* used for both SeqBAIJ and SeqSBAIJ matrices */
1656: PetscErrorCode MatGetRow_SeqBAIJ_private(Mat A, PetscInt row, PetscInt *nz, PetscInt **idx, PetscScalar **v, PetscInt *ai, PetscInt *aj, PetscScalar *aa)
1657: {
1658: PetscInt itmp, i, j, k, M, bn, bp, *idx_i, bs, bs2;
1659: MatScalar *aa_i;
1660: PetscScalar *v_i;
1662: PetscFunctionBegin;
1663: bs = A->rmap->bs;
1664: bs2 = bs * bs;
1665: PetscCheck(row >= 0 && row < A->rmap->N, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Row %" PetscInt_FMT " out of range", row);
1667: bn = row / bs; /* Block number */
1668: bp = row % bs; /* Block Position */
1669: M = ai[bn + 1] - ai[bn];
1670: *nz = bs * M;
1672: if (v) {
1673: *v = NULL;
1674: if (*nz) {
1675: PetscCall(PetscMalloc1(*nz, v));
1676: for (i = 0; i < M; i++) { /* for each block in the block row */
1677: v_i = *v + i * bs;
1678: aa_i = aa + bs2 * (ai[bn] + i);
1679: for (j = bp, k = 0; j < bs2; j += bs, k++) v_i[k] = aa_i[j];
1680: }
1681: }
1682: }
1684: if (idx) {
1685: *idx = NULL;
1686: if (*nz) {
1687: PetscCall(PetscMalloc1(*nz, idx));
1688: for (i = 0; i < M; i++) { /* for each block in the block row */
1689: idx_i = *idx + i * bs;
1690: itmp = bs * aj[ai[bn] + i];
1691: for (j = 0; j < bs; j++) idx_i[j] = itmp++;
1692: }
1693: }
1694: }
1695: PetscFunctionReturn(PETSC_SUCCESS);
1696: }
1698: PetscErrorCode MatGetRow_SeqBAIJ(Mat A, PetscInt row, PetscInt *nz, PetscInt **idx, PetscScalar **v)
1699: {
1700: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
1702: PetscFunctionBegin;
1703: PetscCall(MatGetRow_SeqBAIJ_private(A, row, nz, idx, v, a->i, a->j, a->a));
1704: PetscFunctionReturn(PETSC_SUCCESS);
1705: }
1707: PetscErrorCode MatRestoreRow_SeqBAIJ(Mat A, PetscInt row, PetscInt *nz, PetscInt **idx, PetscScalar **v)
1708: {
1709: PetscFunctionBegin;
1710: if (nz) *nz = 0;
1711: if (idx) PetscCall(PetscFree(*idx));
1712: if (v) PetscCall(PetscFree(*v));
1713: PetscFunctionReturn(PETSC_SUCCESS);
1714: }
1716: PetscErrorCode MatTranspose_SeqBAIJ(Mat A, MatReuse reuse, Mat *B)
1717: {
1718: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data, *at;
1719: Mat C;
1720: PetscInt i, j, k, *aj = a->j, *ai = a->i, bs = A->rmap->bs, mbs = a->mbs, nbs = a->nbs, *atfill;
1721: PetscInt bs2 = a->bs2, *ati, *atj, anzj, kr;
1722: MatScalar *ata, *aa = a->a;
1724: PetscFunctionBegin;
1725: if (reuse == MAT_REUSE_MATRIX) PetscCall(MatTransposeCheckNonzeroState_Private(A, *B));
1726: PetscCall(PetscCalloc1(1 + nbs, &atfill));
1727: if (reuse == MAT_INITIAL_MATRIX || reuse == MAT_INPLACE_MATRIX) {
1728: for (i = 0; i < ai[mbs]; i++) atfill[aj[i]] += 1; /* count num of non-zeros in row aj[i] */
1730: PetscCall(MatCreate(PetscObjectComm((PetscObject)A), &C));
1731: PetscCall(MatSetSizes(C, A->cmap->n, A->rmap->N, A->cmap->n, A->rmap->N));
1732: PetscCall(MatSetType(C, ((PetscObject)A)->type_name));
1733: PetscCall(MatSeqBAIJSetPreallocation(C, bs, 0, atfill));
1735: at = (Mat_SeqBAIJ *)C->data;
1736: ati = at->i;
1737: for (i = 0; i < nbs; i++) at->ilen[i] = at->imax[i] = ati[i + 1] - ati[i];
1738: } else {
1739: C = *B;
1740: at = (Mat_SeqBAIJ *)C->data;
1741: ati = at->i;
1742: }
1744: atj = at->j;
1745: ata = at->a;
1747: /* Copy ati into atfill so we have locations of the next free space in atj */
1748: PetscCall(PetscArraycpy(atfill, ati, nbs));
1750: /* Walk through A row-wise and mark nonzero entries of A^T. */
1751: for (i = 0; i < mbs; i++) {
1752: anzj = ai[i + 1] - ai[i];
1753: for (j = 0; j < anzj; j++) {
1754: atj[atfill[*aj]] = i;
1755: for (kr = 0; kr < bs; kr++) {
1756: for (k = 0; k < bs; k++) ata[bs2 * atfill[*aj] + k * bs + kr] = *aa++;
1757: }
1758: atfill[*aj++] += 1;
1759: }
1760: }
1761: PetscCall(MatAssemblyBegin(C, MAT_FINAL_ASSEMBLY));
1762: PetscCall(MatAssemblyEnd(C, MAT_FINAL_ASSEMBLY));
1764: /* Clean up temporary space and complete requests. */
1765: PetscCall(PetscFree(atfill));
1767: if (reuse == MAT_INITIAL_MATRIX || reuse == MAT_REUSE_MATRIX) {
1768: PetscCall(MatSetBlockSizes(C, PetscAbs(A->cmap->bs), PetscAbs(A->rmap->bs)));
1769: *B = C;
1770: } else {
1771: PetscCall(MatHeaderMerge(A, &C));
1772: }
1773: PetscFunctionReturn(PETSC_SUCCESS);
1774: }
1776: static PetscErrorCode MatIsTranspose_SeqBAIJ(Mat A, Mat B, PetscReal tol, PetscBool *f)
1777: {
1778: Mat Btrans;
1780: PetscFunctionBegin;
1781: *f = PETSC_FALSE;
1782: PetscCall(MatTranspose(A, MAT_INITIAL_MATRIX, &Btrans));
1783: PetscCall(MatEqual_SeqBAIJ(B, Btrans, f));
1784: PetscCall(MatDestroy(&Btrans));
1785: PetscFunctionReturn(PETSC_SUCCESS);
1786: }
1788: /* Used for both SeqBAIJ and SeqSBAIJ matrices */
1789: PetscErrorCode MatView_SeqBAIJ_Binary(Mat mat, PetscViewer viewer)
1790: {
1791: Mat_SeqBAIJ *A = (Mat_SeqBAIJ *)mat->data;
1792: PetscInt header[4], M, N, m, bs, nz, cnt, i, j, k, l;
1793: PetscInt *rowlens, *colidxs;
1794: PetscScalar *matvals;
1796: PetscFunctionBegin;
1797: PetscCall(PetscViewerSetUp(viewer));
1799: M = mat->rmap->N;
1800: N = mat->cmap->N;
1801: m = mat->rmap->n;
1802: bs = mat->rmap->bs;
1803: nz = bs * bs * A->nz;
1805: /* write matrix header */
1806: header[0] = MAT_FILE_CLASSID;
1807: header[1] = M;
1808: header[2] = N;
1809: header[3] = nz;
1810: PetscCall(PetscViewerBinaryWrite(viewer, header, 4, PETSC_INT));
1812: /* store row lengths */
1813: PetscCall(PetscMalloc1(m, &rowlens));
1814: for (cnt = 0, i = 0; i < A->mbs; i++)
1815: for (j = 0; j < bs; j++) rowlens[cnt++] = bs * (A->i[i + 1] - A->i[i]);
1816: PetscCall(PetscViewerBinaryWrite(viewer, rowlens, m, PETSC_INT));
1817: PetscCall(PetscFree(rowlens));
1819: /* store column indices */
1820: PetscCall(PetscMalloc1(nz, &colidxs));
1821: for (cnt = 0, i = 0; i < A->mbs; i++)
1822: for (k = 0; k < bs; k++)
1823: for (j = A->i[i]; j < A->i[i + 1]; j++)
1824: for (l = 0; l < bs; l++) colidxs[cnt++] = bs * A->j[j] + l;
1825: PetscCheck(cnt == nz, PETSC_COMM_SELF, PETSC_ERR_LIB, "Internal PETSc error: cnt = %" PetscInt_FMT " nz = %" PetscInt_FMT, cnt, nz);
1826: PetscCall(PetscViewerBinaryWrite(viewer, colidxs, nz, PETSC_INT));
1827: PetscCall(PetscFree(colidxs));
1829: /* store nonzero values */
1830: PetscCall(PetscMalloc1(nz, &matvals));
1831: for (cnt = 0, i = 0; i < A->mbs; i++)
1832: for (k = 0; k < bs; k++)
1833: for (j = A->i[i]; j < A->i[i + 1]; j++)
1834: for (l = 0; l < bs; l++) matvals[cnt++] = A->a[bs * (bs * j + l) + k];
1835: PetscCheck(cnt == nz, PETSC_COMM_SELF, PETSC_ERR_LIB, "Internal PETSc error: cnt = %" PetscInt_FMT " nz = %" PetscInt_FMT, cnt, nz);
1836: PetscCall(PetscViewerBinaryWrite(viewer, matvals, nz, PETSC_SCALAR));
1837: PetscCall(PetscFree(matvals));
1839: /* write block size option to the viewer's .info file */
1840: PetscCall(MatView_Binary_BlockSizes(mat, viewer));
1841: PetscFunctionReturn(PETSC_SUCCESS);
1842: }
1844: static PetscErrorCode MatView_SeqBAIJ_ASCII_structonly(Mat A, PetscViewer viewer)
1845: {
1846: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
1847: PetscInt i, bs = A->rmap->bs, k;
1849: PetscFunctionBegin;
1850: PetscCall(PetscViewerASCIIUseTabs(viewer, PETSC_FALSE));
1851: for (i = 0; i < a->mbs; i++) {
1852: PetscCall(PetscViewerASCIIPrintf(viewer, "row %" PetscInt_FMT "-%" PetscInt_FMT ":", i * bs, i * bs + bs - 1));
1853: for (k = a->i[i]; k < a->i[i + 1]; k++) PetscCall(PetscViewerASCIIPrintf(viewer, " (%" PetscInt_FMT "-%" PetscInt_FMT ") ", bs * a->j[k], bs * a->j[k] + bs - 1));
1854: PetscCall(PetscViewerASCIIPrintf(viewer, "\n"));
1855: }
1856: PetscCall(PetscViewerASCIIUseTabs(viewer, PETSC_TRUE));
1857: PetscFunctionReturn(PETSC_SUCCESS);
1858: }
1860: static PetscErrorCode MatView_SeqBAIJ_ASCII(Mat A, PetscViewer viewer)
1861: {
1862: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
1863: PetscInt i, j, bs = A->rmap->bs, k, l, bs2 = a->bs2;
1864: PetscViewerFormat format;
1866: PetscFunctionBegin;
1867: if (A->structure_only) {
1868: PetscCall(MatView_SeqBAIJ_ASCII_structonly(A, viewer));
1869: PetscFunctionReturn(PETSC_SUCCESS);
1870: }
1872: PetscCall(PetscViewerGetFormat(viewer, &format));
1873: if (format == PETSC_VIEWER_ASCII_INFO || format == PETSC_VIEWER_ASCII_INFO_DETAIL) {
1874: PetscCall(PetscViewerASCIIPrintf(viewer, " block size is %" PetscInt_FMT "\n", bs));
1875: } else if (format == PETSC_VIEWER_ASCII_MATLAB) {
1876: const char *matname;
1877: Mat aij;
1878: PetscCall(MatConvert(A, MATSEQAIJ, MAT_INITIAL_MATRIX, &aij));
1879: PetscCall(PetscObjectGetName((PetscObject)A, &matname));
1880: PetscCall(PetscObjectSetName((PetscObject)aij, matname));
1881: PetscCall(MatView(aij, viewer));
1882: PetscCall(MatDestroy(&aij));
1883: } else if (format == PETSC_VIEWER_ASCII_FACTOR_INFO) {
1884: PetscFunctionReturn(PETSC_SUCCESS);
1885: } else if (format == PETSC_VIEWER_ASCII_COMMON) {
1886: PetscCall(PetscViewerASCIIUseTabs(viewer, PETSC_FALSE));
1887: for (i = 0; i < a->mbs; i++) {
1888: for (j = 0; j < bs; j++) {
1889: PetscCall(PetscViewerASCIIPrintf(viewer, "row %" PetscInt_FMT ":", i * bs + j));
1890: for (k = a->i[i]; k < a->i[i + 1]; k++) {
1891: for (l = 0; l < bs; l++) {
1892: #if defined(PETSC_USE_COMPLEX)
1893: if (PetscImaginaryPart(a->a[bs2 * k + l * bs + j]) > 0.0 && PetscRealPart(a->a[bs2 * k + l * bs + j]) != 0.0) {
1894: PetscCall(PetscViewerASCIIPrintf(viewer, " (%" PetscInt_FMT ", %g + %gi) ", bs * a->j[k] + l, (double)PetscRealPart(a->a[bs2 * k + l * bs + j]), (double)PetscImaginaryPart(a->a[bs2 * k + l * bs + j])));
1895: } else if (PetscImaginaryPart(a->a[bs2 * k + l * bs + j]) < 0.0 && PetscRealPart(a->a[bs2 * k + l * bs + j]) != 0.0) {
1896: PetscCall(PetscViewerASCIIPrintf(viewer, " (%" PetscInt_FMT ", %g - %gi) ", bs * a->j[k] + l, (double)PetscRealPart(a->a[bs2 * k + l * bs + j]), -(double)PetscImaginaryPart(a->a[bs2 * k + l * bs + j])));
1897: } else if (PetscRealPart(a->a[bs2 * k + l * bs + j]) != 0.0) {
1898: PetscCall(PetscViewerASCIIPrintf(viewer, " (%" PetscInt_FMT ", %g) ", bs * a->j[k] + l, (double)PetscRealPart(a->a[bs2 * k + l * bs + j])));
1899: }
1900: #else
1901: if (a->a[bs2 * k + l * bs + j] != 0.0) PetscCall(PetscViewerASCIIPrintf(viewer, " (%" PetscInt_FMT ", %g) ", bs * a->j[k] + l, (double)a->a[bs2 * k + l * bs + j]));
1902: #endif
1903: }
1904: }
1905: PetscCall(PetscViewerASCIIPrintf(viewer, "\n"));
1906: }
1907: }
1908: PetscCall(PetscViewerASCIIUseTabs(viewer, PETSC_TRUE));
1909: } else {
1910: PetscCall(PetscViewerASCIIUseTabs(viewer, PETSC_FALSE));
1911: for (i = 0; i < a->mbs; i++) {
1912: for (j = 0; j < bs; j++) {
1913: PetscCall(PetscViewerASCIIPrintf(viewer, "row %" PetscInt_FMT ":", i * bs + j));
1914: for (k = a->i[i]; k < a->i[i + 1]; k++) {
1915: for (l = 0; l < bs; l++) {
1916: #if defined(PETSC_USE_COMPLEX)
1917: if (PetscImaginaryPart(a->a[bs2 * k + l * bs + j]) > 0.0) {
1918: PetscCall(PetscViewerASCIIPrintf(viewer, " (%" PetscInt_FMT ", %g + %g i) ", bs * a->j[k] + l, (double)PetscRealPart(a->a[bs2 * k + l * bs + j]), (double)PetscImaginaryPart(a->a[bs2 * k + l * bs + j])));
1919: } else if (PetscImaginaryPart(a->a[bs2 * k + l * bs + j]) < 0.0) {
1920: PetscCall(PetscViewerASCIIPrintf(viewer, " (%" PetscInt_FMT ", %g - %g i) ", bs * a->j[k] + l, (double)PetscRealPart(a->a[bs2 * k + l * bs + j]), -(double)PetscImaginaryPart(a->a[bs2 * k + l * bs + j])));
1921: } else {
1922: PetscCall(PetscViewerASCIIPrintf(viewer, " (%" PetscInt_FMT ", %g) ", bs * a->j[k] + l, (double)PetscRealPart(a->a[bs2 * k + l * bs + j])));
1923: }
1924: #else
1925: PetscCall(PetscViewerASCIIPrintf(viewer, " (%" PetscInt_FMT ", %g) ", bs * a->j[k] + l, (double)a->a[bs2 * k + l * bs + j]));
1926: #endif
1927: }
1928: }
1929: PetscCall(PetscViewerASCIIPrintf(viewer, "\n"));
1930: }
1931: }
1932: PetscCall(PetscViewerASCIIUseTabs(viewer, PETSC_TRUE));
1933: }
1934: PetscCall(PetscViewerFlush(viewer));
1935: PetscFunctionReturn(PETSC_SUCCESS);
1936: }
1938: #include <petscdraw.h>
1939: static PetscErrorCode MatView_SeqBAIJ_Draw_Zoom(PetscDraw draw, void *Aa)
1940: {
1941: Mat A = (Mat)Aa;
1942: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
1943: PetscInt row, i, j, k, l, mbs = a->mbs, color, bs = A->rmap->bs, bs2 = a->bs2;
1944: PetscReal xl, yl, xr, yr, x_l, x_r, y_l, y_r;
1945: MatScalar *aa;
1946: PetscViewer viewer;
1947: PetscViewerFormat format;
1949: PetscFunctionBegin;
1950: PetscCall(PetscObjectQuery((PetscObject)A, "Zoomviewer", (PetscObject *)&viewer));
1951: PetscCall(PetscViewerGetFormat(viewer, &format));
1952: PetscCall(PetscDrawGetCoordinates(draw, &xl, &yl, &xr, &yr));
1954: /* loop over matrix elements drawing boxes */
1956: if (format != PETSC_VIEWER_DRAW_CONTOUR) {
1957: PetscDrawCollectiveBegin(draw);
1958: /* Blue for negative, Cyan for zero and Red for positive */
1959: color = PETSC_DRAW_BLUE;
1960: for (i = 0, row = 0; i < mbs; i++, row += bs) {
1961: for (j = a->i[i]; j < a->i[i + 1]; j++) {
1962: y_l = A->rmap->N - row - 1.0;
1963: y_r = y_l + 1.0;
1964: x_l = a->j[j] * bs;
1965: x_r = x_l + 1.0;
1966: aa = a->a + j * bs2;
1967: for (k = 0; k < bs; k++) {
1968: for (l = 0; l < bs; l++) {
1969: if (PetscRealPart(*aa++) >= 0.) continue;
1970: PetscCall(PetscDrawRectangle(draw, x_l + k, y_l - l, x_r + k, y_r - l, color, color, color, color));
1971: }
1972: }
1973: }
1974: }
1975: color = PETSC_DRAW_CYAN;
1976: for (i = 0, row = 0; i < mbs; i++, row += bs) {
1977: for (j = a->i[i]; j < a->i[i + 1]; j++) {
1978: y_l = A->rmap->N - row - 1.0;
1979: y_r = y_l + 1.0;
1980: x_l = a->j[j] * bs;
1981: x_r = x_l + 1.0;
1982: aa = a->a + j * bs2;
1983: for (k = 0; k < bs; k++) {
1984: for (l = 0; l < bs; l++) {
1985: if (PetscRealPart(*aa++) != 0.) continue;
1986: PetscCall(PetscDrawRectangle(draw, x_l + k, y_l - l, x_r + k, y_r - l, color, color, color, color));
1987: }
1988: }
1989: }
1990: }
1991: color = PETSC_DRAW_RED;
1992: for (i = 0, row = 0; i < mbs; i++, row += bs) {
1993: for (j = a->i[i]; j < a->i[i + 1]; j++) {
1994: y_l = A->rmap->N - row - 1.0;
1995: y_r = y_l + 1.0;
1996: x_l = a->j[j] * bs;
1997: x_r = x_l + 1.0;
1998: aa = a->a + j * bs2;
1999: for (k = 0; k < bs; k++) {
2000: for (l = 0; l < bs; l++) {
2001: if (PetscRealPart(*aa++) <= 0.) continue;
2002: PetscCall(PetscDrawRectangle(draw, x_l + k, y_l - l, x_r + k, y_r - l, color, color, color, color));
2003: }
2004: }
2005: }
2006: }
2007: PetscDrawCollectiveEnd(draw);
2008: } else {
2009: /* use contour shading to indicate magnitude of values */
2010: /* first determine max of all nonzero values */
2011: PetscReal minv = 0.0, maxv = 0.0;
2012: PetscDraw popup;
2014: for (i = 0; i < a->nz * a->bs2; i++) {
2015: if (PetscAbsScalar(a->a[i]) > maxv) maxv = PetscAbsScalar(a->a[i]);
2016: }
2017: if (minv >= maxv) maxv = minv + PETSC_SMALL;
2018: PetscCall(PetscDrawGetPopup(draw, &popup));
2019: PetscCall(PetscDrawScalePopup(popup, 0.0, maxv));
2021: PetscDrawCollectiveBegin(draw);
2022: for (i = 0, row = 0; i < mbs; i++, row += bs) {
2023: for (j = a->i[i]; j < a->i[i + 1]; j++) {
2024: y_l = A->rmap->N - row - 1.0;
2025: y_r = y_l + 1.0;
2026: x_l = a->j[j] * bs;
2027: x_r = x_l + 1.0;
2028: aa = a->a + j * bs2;
2029: for (k = 0; k < bs; k++) {
2030: for (l = 0; l < bs; l++) {
2031: MatScalar v = *aa++;
2032: color = PetscDrawRealToColor(PetscAbsScalar(v), minv, maxv);
2033: PetscCall(PetscDrawRectangle(draw, x_l + k, y_l - l, x_r + k, y_r - l, color, color, color, color));
2034: }
2035: }
2036: }
2037: }
2038: PetscDrawCollectiveEnd(draw);
2039: }
2040: PetscFunctionReturn(PETSC_SUCCESS);
2041: }
2043: static PetscErrorCode MatView_SeqBAIJ_Draw(Mat A, PetscViewer viewer)
2044: {
2045: PetscReal xl, yl, xr, yr, w, h;
2046: PetscDraw draw;
2047: PetscBool isnull;
2049: PetscFunctionBegin;
2050: PetscCall(PetscViewerDrawGetDraw(viewer, 0, &draw));
2051: PetscCall(PetscDrawIsNull(draw, &isnull));
2052: if (isnull) PetscFunctionReturn(PETSC_SUCCESS);
2054: xr = A->cmap->n;
2055: yr = A->rmap->N;
2056: h = yr / 10.0;
2057: w = xr / 10.0;
2058: xr += w;
2059: yr += h;
2060: xl = -w;
2061: yl = -h;
2062: PetscCall(PetscDrawSetCoordinates(draw, xl, yl, xr, yr));
2063: PetscCall(PetscObjectCompose((PetscObject)A, "Zoomviewer", (PetscObject)viewer));
2064: PetscCall(PetscDrawZoom(draw, MatView_SeqBAIJ_Draw_Zoom, A));
2065: PetscCall(PetscObjectCompose((PetscObject)A, "Zoomviewer", NULL));
2066: PetscCall(PetscDrawSave(draw));
2067: PetscFunctionReturn(PETSC_SUCCESS);
2068: }
2070: PetscErrorCode MatView_SeqBAIJ(Mat A, PetscViewer viewer)
2071: {
2072: PetscBool iascii, isbinary, isdraw;
2074: PetscFunctionBegin;
2075: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &iascii));
2076: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERBINARY, &isbinary));
2077: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERDRAW, &isdraw));
2078: if (iascii) {
2079: PetscCall(MatView_SeqBAIJ_ASCII(A, viewer));
2080: } else if (isbinary) {
2081: PetscCall(MatView_SeqBAIJ_Binary(A, viewer));
2082: } else if (isdraw) {
2083: PetscCall(MatView_SeqBAIJ_Draw(A, viewer));
2084: } else {
2085: Mat B;
2086: PetscCall(MatConvert(A, MATSEQAIJ, MAT_INITIAL_MATRIX, &B));
2087: PetscCall(MatView(B, viewer));
2088: PetscCall(MatDestroy(&B));
2089: }
2090: PetscFunctionReturn(PETSC_SUCCESS);
2091: }
2093: PetscErrorCode MatGetValues_SeqBAIJ(Mat A, PetscInt m, const PetscInt im[], PetscInt n, const PetscInt in[], PetscScalar v[])
2094: {
2095: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
2096: PetscInt *rp, k, low, high, t, row, nrow, i, col, l, *aj = a->j;
2097: PetscInt *ai = a->i, *ailen = a->ilen;
2098: PetscInt brow, bcol, ridx, cidx, bs = A->rmap->bs, bs2 = a->bs2;
2099: MatScalar *ap, *aa = a->a;
2101: PetscFunctionBegin;
2102: for (k = 0; k < m; k++) { /* loop over rows */
2103: row = im[k];
2104: brow = row / bs;
2105: if (row < 0) {
2106: v += n;
2107: continue;
2108: } /* negative row */
2109: PetscCheck(row < A->rmap->N, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Row %" PetscInt_FMT " too large", row);
2110: rp = aj ? aj + ai[brow] : NULL; /* mustn't add to NULL, that is UB */
2111: ap = aa ? aa + bs2 * ai[brow] : NULL; /* mustn't add to NULL, that is UB */
2112: nrow = ailen[brow];
2113: for (l = 0; l < n; l++) { /* loop over columns */
2114: if (in[l] < 0) {
2115: v++;
2116: continue;
2117: } /* negative column */
2118: PetscCheck(in[l] < A->cmap->n, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Column %" PetscInt_FMT " too large", in[l]);
2119: col = in[l];
2120: bcol = col / bs;
2121: cidx = col % bs;
2122: ridx = row % bs;
2123: high = nrow;
2124: low = 0; /* assume unsorted */
2125: while (high - low > 5) {
2126: t = (low + high) / 2;
2127: if (rp[t] > bcol) high = t;
2128: else low = t;
2129: }
2130: for (i = low; i < high; i++) {
2131: if (rp[i] > bcol) break;
2132: if (rp[i] == bcol) {
2133: *v++ = ap[bs2 * i + bs * cidx + ridx];
2134: goto finished;
2135: }
2136: }
2137: *v++ = 0.0;
2138: finished:;
2139: }
2140: }
2141: PetscFunctionReturn(PETSC_SUCCESS);
2142: }
2144: PetscErrorCode MatSetValuesBlocked_SeqBAIJ(Mat A, PetscInt m, const PetscInt im[], PetscInt n, const PetscInt in[], const PetscScalar v[], InsertMode is)
2145: {
2146: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
2147: PetscInt *rp, k, low, high, t, ii, jj, row, nrow, i, col, l, rmax, N, lastcol = -1;
2148: PetscInt *imax = a->imax, *ai = a->i, *ailen = a->ilen;
2149: PetscInt *aj = a->j, nonew = a->nonew, bs2 = a->bs2, bs = A->rmap->bs, stepval;
2150: PetscBool roworiented = a->roworiented;
2151: const PetscScalar *value = v;
2152: MatScalar *ap = NULL, *aa = a->a, *bap;
2154: PetscFunctionBegin;
2155: if (roworiented) {
2156: stepval = (n - 1) * bs;
2157: } else {
2158: stepval = (m - 1) * bs;
2159: }
2160: for (k = 0; k < m; k++) { /* loop over added rows */
2161: row = im[k];
2162: if (row < 0) continue;
2163: PetscCheck(row < a->mbs, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Block row index too large %" PetscInt_FMT " max %" PetscInt_FMT, row, a->mbs - 1);
2164: rp = aj + ai[row];
2165: if (!A->structure_only) ap = aa + bs2 * ai[row];
2166: rmax = imax[row];
2167: nrow = ailen[row];
2168: low = 0;
2169: high = nrow;
2170: for (l = 0; l < n; l++) { /* loop over added columns */
2171: if (in[l] < 0) continue;
2172: PetscCheck(in[l] < a->nbs, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Block column index too large %" PetscInt_FMT " max %" PetscInt_FMT, in[l], a->nbs - 1);
2173: col = in[l];
2174: if (!A->structure_only) {
2175: if (roworiented) {
2176: value = v + (k * (stepval + bs) + l) * bs;
2177: } else {
2178: value = v + (l * (stepval + bs) + k) * bs;
2179: }
2180: }
2181: if (col <= lastcol) low = 0;
2182: else high = nrow;
2183: lastcol = col;
2184: while (high - low > 7) {
2185: t = (low + high) / 2;
2186: if (rp[t] > col) high = t;
2187: else low = t;
2188: }
2189: for (i = low; i < high; i++) {
2190: if (rp[i] > col) break;
2191: if (rp[i] == col) {
2192: if (A->structure_only) goto noinsert2;
2193: bap = ap + bs2 * i;
2194: if (roworiented) {
2195: if (is == ADD_VALUES) {
2196: for (ii = 0; ii < bs; ii++, value += stepval) {
2197: for (jj = ii; jj < bs2; jj += bs) bap[jj] += *value++;
2198: }
2199: } else {
2200: for (ii = 0; ii < bs; ii++, value += stepval) {
2201: for (jj = ii; jj < bs2; jj += bs) bap[jj] = *value++;
2202: }
2203: }
2204: } else {
2205: if (is == ADD_VALUES) {
2206: for (ii = 0; ii < bs; ii++, value += bs + stepval) {
2207: for (jj = 0; jj < bs; jj++) bap[jj] += value[jj];
2208: bap += bs;
2209: }
2210: } else {
2211: for (ii = 0; ii < bs; ii++, value += bs + stepval) {
2212: for (jj = 0; jj < bs; jj++) bap[jj] = value[jj];
2213: bap += bs;
2214: }
2215: }
2216: }
2217: goto noinsert2;
2218: }
2219: }
2220: if (nonew == 1) goto noinsert2;
2221: PetscCheck(nonew != -1, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Inserting a new blocked index new nonzero block (%" PetscInt_FMT ", %" PetscInt_FMT ") in the matrix", row, col);
2222: if (A->structure_only) {
2223: MatSeqXAIJReallocateAIJ_structure_only(A, a->mbs, bs2, nrow, row, col, rmax, ai, aj, rp, imax, nonew, MatScalar);
2224: } else {
2225: MatSeqXAIJReallocateAIJ(A, a->mbs, bs2, nrow, row, col, rmax, aa, ai, aj, rp, ap, imax, nonew, MatScalar);
2226: }
2227: N = nrow++ - 1;
2228: high++;
2229: /* shift up all the later entries in this row */
2230: PetscCall(PetscArraymove(rp + i + 1, rp + i, N - i + 1));
2231: rp[i] = col;
2232: if (!A->structure_only) {
2233: PetscCall(PetscArraymove(ap + bs2 * (i + 1), ap + bs2 * i, bs2 * (N - i + 1)));
2234: bap = ap + bs2 * i;
2235: if (roworiented) {
2236: for (ii = 0; ii < bs; ii++, value += stepval) {
2237: for (jj = ii; jj < bs2; jj += bs) bap[jj] = *value++;
2238: }
2239: } else {
2240: for (ii = 0; ii < bs; ii++, value += stepval) {
2241: for (jj = 0; jj < bs; jj++) *bap++ = *value++;
2242: }
2243: }
2244: }
2245: noinsert2:;
2246: low = i;
2247: }
2248: ailen[row] = nrow;
2249: }
2250: PetscFunctionReturn(PETSC_SUCCESS);
2251: }
2253: PetscErrorCode MatAssemblyEnd_SeqBAIJ(Mat A, MatAssemblyType mode)
2254: {
2255: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
2256: PetscInt fshift = 0, i, *ai = a->i, *aj = a->j, *imax = a->imax;
2257: PetscInt m = A->rmap->N, *ip, N, *ailen = a->ilen;
2258: PetscInt mbs = a->mbs, bs2 = a->bs2, rmax = 0;
2259: MatScalar *aa = a->a, *ap;
2260: PetscReal ratio = 0.6;
2262: PetscFunctionBegin;
2263: if (mode == MAT_FLUSH_ASSEMBLY) PetscFunctionReturn(PETSC_SUCCESS);
2265: if (m) rmax = ailen[0];
2266: for (i = 1; i < mbs; i++) {
2267: /* move each row back by the amount of empty slots (fshift) before it*/
2268: fshift += imax[i - 1] - ailen[i - 1];
2269: rmax = PetscMax(rmax, ailen[i]);
2270: if (fshift) {
2271: ip = aj + ai[i];
2272: ap = aa + bs2 * ai[i];
2273: N = ailen[i];
2274: PetscCall(PetscArraymove(ip - fshift, ip, N));
2275: if (!A->structure_only) PetscCall(PetscArraymove(ap - bs2 * fshift, ap, bs2 * N));
2276: }
2277: ai[i] = ai[i - 1] + ailen[i - 1];
2278: }
2279: if (mbs) {
2280: fshift += imax[mbs - 1] - ailen[mbs - 1];
2281: ai[mbs] = ai[mbs - 1] + ailen[mbs - 1];
2282: }
2284: /* reset ilen and imax for each row */
2285: a->nonzerorowcnt = 0;
2286: if (A->structure_only) {
2287: PetscCall(PetscFree2(a->imax, a->ilen));
2288: } else { /* !A->structure_only */
2289: for (i = 0; i < mbs; i++) {
2290: ailen[i] = imax[i] = ai[i + 1] - ai[i];
2291: a->nonzerorowcnt += ((ai[i + 1] - ai[i]) > 0);
2292: }
2293: }
2294: a->nz = ai[mbs];
2296: /* diagonals may have moved, so kill the diagonal pointers */
2297: a->idiagvalid = PETSC_FALSE;
2298: if (fshift && a->diag) PetscCall(PetscFree(a->diag));
2299: if (fshift) PetscCheck(a->nounused != -1, PETSC_COMM_SELF, PETSC_ERR_PLIB, "Unused space detected in matrix: %" PetscInt_FMT " X %" PetscInt_FMT " block size %" PetscInt_FMT ", %" PetscInt_FMT " unneeded", m, A->cmap->n, A->rmap->bs, fshift * bs2);
2300: PetscCall(PetscInfo(A, "Matrix size: %" PetscInt_FMT " X %" PetscInt_FMT ", block size %" PetscInt_FMT "; storage space: %" PetscInt_FMT " unneeded, %" PetscInt_FMT " used\n", m, A->cmap->n, A->rmap->bs, fshift * bs2, a->nz * bs2));
2301: PetscCall(PetscInfo(A, "Number of mallocs during MatSetValues is %" PetscInt_FMT "\n", a->reallocs));
2302: PetscCall(PetscInfo(A, "Most nonzeros blocks in any row is %" PetscInt_FMT "\n", rmax));
2304: A->info.mallocs += a->reallocs;
2305: a->reallocs = 0;
2306: A->info.nz_unneeded = (PetscReal)fshift * bs2;
2307: a->rmax = rmax;
2309: if (!A->structure_only) PetscCall(MatCheckCompressedRow(A, a->nonzerorowcnt, &a->compressedrow, a->i, mbs, ratio));
2310: PetscFunctionReturn(PETSC_SUCCESS);
2311: }
2313: /*
2314: This function returns an array of flags which indicate the locations of contiguous
2315: blocks that should be zeroed. for eg: if bs = 3 and is = [0,1,2,3,5,6,7,8,9]
2316: then the resulting sizes = [3,1,1,3,1] corresponding to sets [(0,1,2),(3),(5),(6,7,8),(9)]
2317: Assume: sizes should be long enough to hold all the values.
2318: */
2319: static PetscErrorCode MatZeroRows_SeqBAIJ_Check_Blocks(PetscInt idx[], PetscInt n, PetscInt bs, PetscInt sizes[], PetscInt *bs_max)
2320: {
2321: PetscInt j = 0;
2323: PetscFunctionBegin;
2324: for (PetscInt i = 0; i < n; j++) {
2325: PetscInt row = idx[i];
2326: if (row % bs != 0) { /* Not the beginning of a block */
2327: sizes[j] = 1;
2328: i++;
2329: } else if (i + bs > n) { /* complete block doesn't exist (at idx end) */
2330: sizes[j] = 1; /* Also makes sure at least 'bs' values exist for next else */
2331: i++;
2332: } else { /* Beginning of the block, so check if the complete block exists */
2333: PetscBool flg = PETSC_TRUE;
2334: for (PetscInt k = 1; k < bs; k++) {
2335: if (row + k != idx[i + k]) { /* break in the block */
2336: flg = PETSC_FALSE;
2337: break;
2338: }
2339: }
2340: if (flg) { /* No break in the bs */
2341: sizes[j] = bs;
2342: i += bs;
2343: } else {
2344: sizes[j] = 1;
2345: i++;
2346: }
2347: }
2348: }
2349: *bs_max = j;
2350: PetscFunctionReturn(PETSC_SUCCESS);
2351: }
2353: PetscErrorCode MatZeroRows_SeqBAIJ(Mat A, PetscInt is_n, const PetscInt is_idx[], PetscScalar diag, Vec x, Vec b)
2354: {
2355: Mat_SeqBAIJ *baij = (Mat_SeqBAIJ *)A->data;
2356: PetscInt i, j, k, count, *rows;
2357: PetscInt bs = A->rmap->bs, bs2 = baij->bs2, *sizes, row, bs_max;
2358: PetscScalar zero = 0.0;
2359: MatScalar *aa;
2360: const PetscScalar *xx;
2361: PetscScalar *bb;
2363: PetscFunctionBegin;
2364: /* fix right hand side if needed */
2365: if (x && b) {
2366: PetscCall(VecGetArrayRead(x, &xx));
2367: PetscCall(VecGetArray(b, &bb));
2368: for (i = 0; i < is_n; i++) bb[is_idx[i]] = diag * xx[is_idx[i]];
2369: PetscCall(VecRestoreArrayRead(x, &xx));
2370: PetscCall(VecRestoreArray(b, &bb));
2371: }
2373: /* Make a copy of the IS and sort it */
2374: /* allocate memory for rows,sizes */
2375: PetscCall(PetscMalloc2(is_n, &rows, 2 * is_n, &sizes));
2377: /* copy IS values to rows, and sort them */
2378: for (i = 0; i < is_n; i++) rows[i] = is_idx[i];
2379: PetscCall(PetscSortInt(is_n, rows));
2381: if (baij->keepnonzeropattern) {
2382: for (i = 0; i < is_n; i++) sizes[i] = 1;
2383: bs_max = is_n;
2384: } else {
2385: PetscCall(MatZeroRows_SeqBAIJ_Check_Blocks(rows, is_n, bs, sizes, &bs_max));
2386: A->nonzerostate++;
2387: }
2389: for (i = 0, j = 0; i < bs_max; j += sizes[i], i++) {
2390: row = rows[j];
2391: PetscCheck(row >= 0 && row <= A->rmap->N, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "row %" PetscInt_FMT " out of range", row);
2392: count = (baij->i[row / bs + 1] - baij->i[row / bs]) * bs;
2393: aa = ((MatScalar *)(baij->a)) + baij->i[row / bs] * bs2 + (row % bs);
2394: if (sizes[i] == bs && !baij->keepnonzeropattern) {
2395: if (diag != (PetscScalar)0.0) {
2396: if (baij->ilen[row / bs] > 0) {
2397: baij->ilen[row / bs] = 1;
2398: baij->j[baij->i[row / bs]] = row / bs;
2400: PetscCall(PetscArrayzero(aa, count * bs));
2401: }
2402: /* Now insert all the diagonal values for this bs */
2403: for (k = 0; k < bs; k++) PetscCall((*A->ops->setvalues)(A, 1, rows + j + k, 1, rows + j + k, &diag, INSERT_VALUES));
2404: } else { /* (diag == 0.0) */
2405: baij->ilen[row / bs] = 0;
2406: } /* end (diag == 0.0) */
2407: } else { /* (sizes[i] != bs) */
2408: PetscAssert(sizes[i] == 1, PETSC_COMM_SELF, PETSC_ERR_PLIB, "Internal Error. Value should be 1");
2409: for (k = 0; k < count; k++) {
2410: aa[0] = zero;
2411: aa += bs;
2412: }
2413: if (diag != (PetscScalar)0.0) PetscCall((*A->ops->setvalues)(A, 1, rows + j, 1, rows + j, &diag, INSERT_VALUES));
2414: }
2415: }
2417: PetscCall(PetscFree2(rows, sizes));
2418: PetscCall(MatAssemblyEnd_SeqBAIJ(A, MAT_FINAL_ASSEMBLY));
2419: PetscFunctionReturn(PETSC_SUCCESS);
2420: }
2422: static PetscErrorCode MatZeroRowsColumns_SeqBAIJ(Mat A, PetscInt is_n, const PetscInt is_idx[], PetscScalar diag, Vec x, Vec b)
2423: {
2424: Mat_SeqBAIJ *baij = (Mat_SeqBAIJ *)A->data;
2425: PetscInt i, j, k, count;
2426: PetscInt bs = A->rmap->bs, bs2 = baij->bs2, row, col;
2427: PetscScalar zero = 0.0;
2428: MatScalar *aa;
2429: const PetscScalar *xx;
2430: PetscScalar *bb;
2431: PetscBool *zeroed, vecs = PETSC_FALSE;
2433: PetscFunctionBegin;
2434: /* fix right hand side if needed */
2435: if (x && b) {
2436: PetscCall(VecGetArrayRead(x, &xx));
2437: PetscCall(VecGetArray(b, &bb));
2438: vecs = PETSC_TRUE;
2439: }
2441: /* zero the columns */
2442: PetscCall(PetscCalloc1(A->rmap->n, &zeroed));
2443: for (i = 0; i < is_n; i++) {
2444: PetscCheck(is_idx[i] >= 0 && is_idx[i] < A->rmap->N, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "row %" PetscInt_FMT " out of range", is_idx[i]);
2445: zeroed[is_idx[i]] = PETSC_TRUE;
2446: }
2447: for (i = 0; i < A->rmap->N; i++) {
2448: if (!zeroed[i]) {
2449: row = i / bs;
2450: for (j = baij->i[row]; j < baij->i[row + 1]; j++) {
2451: for (k = 0; k < bs; k++) {
2452: col = bs * baij->j[j] + k;
2453: if (zeroed[col]) {
2454: aa = ((MatScalar *)(baij->a)) + j * bs2 + (i % bs) + bs * k;
2455: if (vecs) bb[i] -= aa[0] * xx[col];
2456: aa[0] = 0.0;
2457: }
2458: }
2459: }
2460: } else if (vecs) bb[i] = diag * xx[i];
2461: }
2462: PetscCall(PetscFree(zeroed));
2463: if (vecs) {
2464: PetscCall(VecRestoreArrayRead(x, &xx));
2465: PetscCall(VecRestoreArray(b, &bb));
2466: }
2468: /* zero the rows */
2469: for (i = 0; i < is_n; i++) {
2470: row = is_idx[i];
2471: count = (baij->i[row / bs + 1] - baij->i[row / bs]) * bs;
2472: aa = ((MatScalar *)(baij->a)) + baij->i[row / bs] * bs2 + (row % bs);
2473: for (k = 0; k < count; k++) {
2474: aa[0] = zero;
2475: aa += bs;
2476: }
2477: if (diag != (PetscScalar)0.0) PetscUseTypeMethod(A, setvalues, 1, &row, 1, &row, &diag, INSERT_VALUES);
2478: }
2479: PetscCall(MatAssemblyEnd_SeqBAIJ(A, MAT_FINAL_ASSEMBLY));
2480: PetscFunctionReturn(PETSC_SUCCESS);
2481: }
2483: PetscErrorCode MatSetValues_SeqBAIJ(Mat A, PetscInt m, const PetscInt im[], PetscInt n, const PetscInt in[], const PetscScalar v[], InsertMode is)
2484: {
2485: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
2486: PetscInt *rp, k, low, high, t, ii, row, nrow, i, col, l, rmax, N, lastcol = -1;
2487: PetscInt *imax = a->imax, *ai = a->i, *ailen = a->ilen;
2488: PetscInt *aj = a->j, nonew = a->nonew, bs = A->rmap->bs, brow, bcol;
2489: PetscInt ridx, cidx, bs2 = a->bs2;
2490: PetscBool roworiented = a->roworiented;
2491: MatScalar *ap = NULL, value = 0.0, *aa = a->a, *bap;
2493: PetscFunctionBegin;
2494: for (k = 0; k < m; k++) { /* loop over added rows */
2495: row = im[k];
2496: brow = row / bs;
2497: if (row < 0) continue;
2498: PetscCheck(row < A->rmap->N, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Row too large: row %" PetscInt_FMT " max %" PetscInt_FMT, row, A->rmap->N - 1);
2499: rp = aj + ai[brow];
2500: if (!A->structure_only) ap = aa + bs2 * ai[brow];
2501: rmax = imax[brow];
2502: nrow = ailen[brow];
2503: low = 0;
2504: high = nrow;
2505: for (l = 0; l < n; l++) { /* loop over added columns */
2506: if (in[l] < 0) continue;
2507: PetscCheck(in[l] < A->cmap->n, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Column too large: col %" PetscInt_FMT " max %" PetscInt_FMT, in[l], A->cmap->n - 1);
2508: col = in[l];
2509: bcol = col / bs;
2510: ridx = row % bs;
2511: cidx = col % bs;
2512: if (!A->structure_only) {
2513: if (roworiented) {
2514: value = v[l + k * n];
2515: } else {
2516: value = v[k + l * m];
2517: }
2518: }
2519: if (col <= lastcol) low = 0;
2520: else high = nrow;
2521: lastcol = col;
2522: while (high - low > 7) {
2523: t = (low + high) / 2;
2524: if (rp[t] > bcol) high = t;
2525: else low = t;
2526: }
2527: for (i = low; i < high; i++) {
2528: if (rp[i] > bcol) break;
2529: if (rp[i] == bcol) {
2530: bap = ap + bs2 * i + bs * cidx + ridx;
2531: if (!A->structure_only) {
2532: if (is == ADD_VALUES) *bap += value;
2533: else *bap = value;
2534: }
2535: goto noinsert1;
2536: }
2537: }
2538: if (nonew == 1) goto noinsert1;
2539: PetscCheck(nonew != -1, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Inserting a new nonzero (%" PetscInt_FMT ", %" PetscInt_FMT ") in the matrix", row, col);
2540: if (A->structure_only) {
2541: MatSeqXAIJReallocateAIJ_structure_only(A, a->mbs, bs2, nrow, brow, bcol, rmax, ai, aj, rp, imax, nonew, MatScalar);
2542: } else {
2543: MatSeqXAIJReallocateAIJ(A, a->mbs, bs2, nrow, brow, bcol, rmax, aa, ai, aj, rp, ap, imax, nonew, MatScalar);
2544: }
2545: N = nrow++ - 1;
2546: high++;
2547: /* shift up all the later entries in this row */
2548: PetscCall(PetscArraymove(rp + i + 1, rp + i, N - i + 1));
2549: rp[i] = bcol;
2550: if (!A->structure_only) {
2551: PetscCall(PetscArraymove(ap + bs2 * (i + 1), ap + bs2 * i, bs2 * (N - i + 1)));
2552: PetscCall(PetscArrayzero(ap + bs2 * i, bs2));
2553: ap[bs2 * i + bs * cidx + ridx] = value;
2554: }
2555: a->nz++;
2556: A->nonzerostate++;
2557: noinsert1:;
2558: low = i;
2559: }
2560: ailen[brow] = nrow;
2561: }
2562: PetscFunctionReturn(PETSC_SUCCESS);
2563: }
2565: static PetscErrorCode MatILUFactor_SeqBAIJ(Mat inA, IS row, IS col, const MatFactorInfo *info)
2566: {
2567: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)inA->data;
2568: Mat outA;
2569: PetscBool row_identity, col_identity;
2571: PetscFunctionBegin;
2572: PetscCheck(info->levels == 0, PETSC_COMM_SELF, PETSC_ERR_SUP, "Only levels = 0 supported for in-place ILU");
2573: PetscCall(ISIdentity(row, &row_identity));
2574: PetscCall(ISIdentity(col, &col_identity));
2575: PetscCheck(row_identity && col_identity, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "Row and column permutations must be identity for in-place ILU");
2577: outA = inA;
2578: inA->factortype = MAT_FACTOR_LU;
2579: PetscCall(PetscFree(inA->solvertype));
2580: PetscCall(PetscStrallocpy(MATSOLVERPETSC, &inA->solvertype));
2582: PetscCall(MatMarkDiagonal_SeqBAIJ(inA));
2584: PetscCall(PetscObjectReference((PetscObject)row));
2585: PetscCall(ISDestroy(&a->row));
2586: a->row = row;
2587: PetscCall(PetscObjectReference((PetscObject)col));
2588: PetscCall(ISDestroy(&a->col));
2589: a->col = col;
2591: /* Create the invert permutation so that it can be used in MatLUFactorNumeric() */
2592: PetscCall(ISDestroy(&a->icol));
2593: PetscCall(ISInvertPermutation(col, PETSC_DECIDE, &a->icol));
2595: PetscCall(MatSeqBAIJSetNumericFactorization_inplace(inA, (PetscBool)(row_identity && col_identity)));
2596: if (!a->solve_work) PetscCall(PetscMalloc1(inA->rmap->N + inA->rmap->bs, &a->solve_work));
2597: PetscCall(MatLUFactorNumeric(outA, inA, info));
2598: PetscFunctionReturn(PETSC_SUCCESS);
2599: }
2601: static PetscErrorCode MatSeqBAIJSetColumnIndices_SeqBAIJ(Mat mat, const PetscInt *indices)
2602: {
2603: Mat_SeqBAIJ *baij = (Mat_SeqBAIJ *)mat->data;
2605: PetscFunctionBegin;
2606: baij->nz = baij->maxnz;
2607: PetscCall(PetscArraycpy(baij->j, indices, baij->nz));
2608: PetscCall(PetscArraycpy(baij->ilen, baij->imax, baij->mbs));
2609: PetscFunctionReturn(PETSC_SUCCESS);
2610: }
2612: /*@
2613: MatSeqBAIJSetColumnIndices - Set the column indices for all the rows in the matrix.
2615: Input Parameters:
2616: + mat - the `MATSEQBAIJ` matrix
2617: - indices - the column indices
2619: Level: advanced
2621: Notes:
2622: This can be called if you have precomputed the nonzero structure of the
2623: matrix and want to provide it to the matrix object to improve the performance
2624: of the `MatSetValues()` operation.
2626: You MUST have set the correct numbers of nonzeros per row in the call to
2627: `MatCreateSeqBAIJ()`, and the columns indices MUST be sorted.
2629: MUST be called before any calls to `MatSetValues()`
2631: .seealso: [](ch_matrices), `Mat`, `MATSEQBAIJ`, `MatSetValues()`
2632: @*/
2633: PetscErrorCode MatSeqBAIJSetColumnIndices(Mat mat, PetscInt *indices)
2634: {
2635: PetscFunctionBegin;
2638: PetscUseMethod(mat, "MatSeqBAIJSetColumnIndices_C", (Mat, PetscInt *), (mat, indices));
2639: PetscFunctionReturn(PETSC_SUCCESS);
2640: }
2642: PetscErrorCode MatGetRowMaxAbs_SeqBAIJ(Mat A, Vec v, PetscInt idx[])
2643: {
2644: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
2645: PetscInt i, j, n, row, bs, *ai, *aj, mbs;
2646: PetscReal atmp;
2647: PetscScalar *x, zero = 0.0;
2648: MatScalar *aa;
2649: PetscInt ncols, brow, krow, kcol;
2651: PetscFunctionBegin;
2652: /* why is this not a macro???????????????????????????????????????????????????????????????? */
2653: PetscCheck(!A->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2654: bs = A->rmap->bs;
2655: aa = a->a;
2656: ai = a->i;
2657: aj = a->j;
2658: mbs = a->mbs;
2660: PetscCall(VecSet(v, zero));
2661: PetscCall(VecGetArray(v, &x));
2662: PetscCall(VecGetLocalSize(v, &n));
2663: PetscCheck(n == A->rmap->N, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Nonconforming matrix and vector");
2664: for (i = 0; i < mbs; i++) {
2665: ncols = ai[1] - ai[0];
2666: ai++;
2667: brow = bs * i;
2668: for (j = 0; j < ncols; j++) {
2669: for (kcol = 0; kcol < bs; kcol++) {
2670: for (krow = 0; krow < bs; krow++) {
2671: atmp = PetscAbsScalar(*aa);
2672: aa++;
2673: row = brow + krow; /* row index */
2674: if (PetscAbsScalar(x[row]) < atmp) {
2675: x[row] = atmp;
2676: if (idx) idx[row] = bs * (*aj) + kcol;
2677: }
2678: }
2679: }
2680: aj++;
2681: }
2682: }
2683: PetscCall(VecRestoreArray(v, &x));
2684: PetscFunctionReturn(PETSC_SUCCESS);
2685: }
2687: PetscErrorCode MatCopy_SeqBAIJ(Mat A, Mat B, MatStructure str)
2688: {
2689: PetscFunctionBegin;
2690: /* If the two matrices have the same copy implementation, use fast copy. */
2691: if (str == SAME_NONZERO_PATTERN && (A->ops->copy == B->ops->copy)) {
2692: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
2693: Mat_SeqBAIJ *b = (Mat_SeqBAIJ *)B->data;
2694: PetscInt ambs = a->mbs, bmbs = b->mbs, abs = A->rmap->bs, bbs = B->rmap->bs, bs2 = abs * abs;
2696: PetscCheck(a->i[ambs] == b->i[bmbs], PETSC_COMM_SELF, PETSC_ERR_ARG_INCOMP, "Number of nonzero blocks in matrices A %" PetscInt_FMT " and B %" PetscInt_FMT " are different", a->i[ambs], b->i[bmbs]);
2697: PetscCheck(abs == bbs, PETSC_COMM_SELF, PETSC_ERR_ARG_INCOMP, "Block size A %" PetscInt_FMT " and B %" PetscInt_FMT " are different", abs, bbs);
2698: PetscCall(PetscArraycpy(b->a, a->a, bs2 * a->i[ambs]));
2699: PetscCall(PetscObjectStateIncrease((PetscObject)B));
2700: } else {
2701: PetscCall(MatCopy_Basic(A, B, str));
2702: }
2703: PetscFunctionReturn(PETSC_SUCCESS);
2704: }
2706: static PetscErrorCode MatSeqBAIJGetArray_SeqBAIJ(Mat A, PetscScalar *array[])
2707: {
2708: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
2710: PetscFunctionBegin;
2711: *array = a->a;
2712: PetscFunctionReturn(PETSC_SUCCESS);
2713: }
2715: static PetscErrorCode MatSeqBAIJRestoreArray_SeqBAIJ(Mat A, PetscScalar *array[])
2716: {
2717: PetscFunctionBegin;
2718: *array = NULL;
2719: PetscFunctionReturn(PETSC_SUCCESS);
2720: }
2722: PetscErrorCode MatAXPYGetPreallocation_SeqBAIJ(Mat Y, Mat X, PetscInt *nnz)
2723: {
2724: PetscInt bs = Y->rmap->bs, mbs = Y->rmap->N / bs;
2725: Mat_SeqBAIJ *x = (Mat_SeqBAIJ *)X->data;
2726: Mat_SeqBAIJ *y = (Mat_SeqBAIJ *)Y->data;
2728: PetscFunctionBegin;
2729: /* Set the number of nonzeros in the new matrix */
2730: PetscCall(MatAXPYGetPreallocation_SeqX_private(mbs, x->i, x->j, y->i, y->j, nnz));
2731: PetscFunctionReturn(PETSC_SUCCESS);
2732: }
2734: PetscErrorCode MatAXPY_SeqBAIJ(Mat Y, PetscScalar a, Mat X, MatStructure str)
2735: {
2736: Mat_SeqBAIJ *x = (Mat_SeqBAIJ *)X->data, *y = (Mat_SeqBAIJ *)Y->data;
2737: PetscInt bs = Y->rmap->bs, bs2 = bs * bs;
2738: PetscBLASInt one = 1;
2740: PetscFunctionBegin;
2741: if (str == UNKNOWN_NONZERO_PATTERN || (PetscDefined(USE_DEBUG) && str == SAME_NONZERO_PATTERN)) {
2742: PetscBool e = x->nz == y->nz && x->mbs == y->mbs && bs == X->rmap->bs ? PETSC_TRUE : PETSC_FALSE;
2743: if (e) {
2744: PetscCall(PetscArraycmp(x->i, y->i, x->mbs + 1, &e));
2745: if (e) {
2746: PetscCall(PetscArraycmp(x->j, y->j, x->i[x->mbs], &e));
2747: if (e) str = SAME_NONZERO_PATTERN;
2748: }
2749: }
2750: if (!e) PetscCheck(str != SAME_NONZERO_PATTERN, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "MatStructure is not SAME_NONZERO_PATTERN");
2751: }
2752: if (str == SAME_NONZERO_PATTERN) {
2753: PetscScalar alpha = a;
2754: PetscBLASInt bnz;
2755: PetscCall(PetscBLASIntCast(x->nz * bs2, &bnz));
2756: PetscCallBLAS("BLASaxpy", BLASaxpy_(&bnz, &alpha, x->a, &one, y->a, &one));
2757: PetscCall(PetscObjectStateIncrease((PetscObject)Y));
2758: } else if (str == SUBSET_NONZERO_PATTERN) { /* nonzeros of X is a subset of Y's */
2759: PetscCall(MatAXPY_Basic(Y, a, X, str));
2760: } else {
2761: Mat B;
2762: PetscInt *nnz;
2763: PetscCheck(bs == X->rmap->bs, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Matrices must have same block size");
2764: PetscCall(PetscMalloc1(Y->rmap->N, &nnz));
2765: PetscCall(MatCreate(PetscObjectComm((PetscObject)Y), &B));
2766: PetscCall(PetscObjectSetName((PetscObject)B, ((PetscObject)Y)->name));
2767: PetscCall(MatSetSizes(B, Y->rmap->n, Y->cmap->n, Y->rmap->N, Y->cmap->N));
2768: PetscCall(MatSetBlockSizesFromMats(B, Y, Y));
2769: PetscCall(MatSetType(B, (MatType)((PetscObject)Y)->type_name));
2770: PetscCall(MatAXPYGetPreallocation_SeqBAIJ(Y, X, nnz));
2771: PetscCall(MatSeqBAIJSetPreallocation(B, bs, 0, nnz));
2772: PetscCall(MatAXPY_BasicWithPreallocation(B, Y, a, X, str));
2773: PetscCall(MatHeaderMerge(Y, &B));
2774: PetscCall(PetscFree(nnz));
2775: }
2776: PetscFunctionReturn(PETSC_SUCCESS);
2777: }
2779: PETSC_INTERN PetscErrorCode MatConjugate_SeqBAIJ(Mat A)
2780: {
2781: #if PetscDefined(USE_COMPLEX)
2782: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
2783: PetscInt i, nz = a->bs2 * a->i[a->mbs];
2784: MatScalar *aa = a->a;
2786: PetscFunctionBegin;
2787: for (i = 0; i < nz; i++) aa[i] = PetscConj(aa[i]);
2788: PetscFunctionReturn(PETSC_SUCCESS);
2789: #else
2790: (void)A;
2791: return PETSC_SUCCESS;
2792: #endif
2793: }
2795: static PetscErrorCode MatRealPart_SeqBAIJ(Mat A)
2796: {
2797: #if PetscDefined(USE_COMPLEX)
2798: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
2799: PetscInt i, nz = a->bs2 * a->i[a->mbs];
2800: MatScalar *aa = a->a;
2802: PetscFunctionBegin;
2803: for (i = 0; i < nz; i++) aa[i] = PetscRealPart(aa[i]);
2804: PetscFunctionReturn(PETSC_SUCCESS);
2805: #else
2806: (void)A;
2807: return PETSC_SUCCESS;
2808: #endif
2809: }
2811: static PetscErrorCode MatImaginaryPart_SeqBAIJ(Mat A)
2812: {
2813: #if PetscDefined(USE_COMPLEX)
2814: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
2815: PetscInt i, nz = a->bs2 * a->i[a->mbs];
2816: MatScalar *aa = a->a;
2818: PetscFunctionBegin;
2819: for (i = 0; i < nz; i++) aa[i] = PetscImaginaryPart(aa[i]);
2820: PetscFunctionReturn(PETSC_SUCCESS);
2821: #else
2822: (void)A;
2823: return PETSC_SUCCESS;
2824: #endif
2825: }
2827: /*
2828: Code almost identical to MatGetColumnIJ_SeqAIJ() should share common code
2829: */
2830: static PetscErrorCode MatGetColumnIJ_SeqBAIJ(Mat A, PetscInt oshift, PetscBool symmetric, PetscBool inodecompressed, PetscInt *nn, const PetscInt *ia[], const PetscInt *ja[], PetscBool *done)
2831: {
2832: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
2833: PetscInt bs = A->rmap->bs, i, *collengths, *cia, *cja, n = A->cmap->n / bs, m = A->rmap->n / bs;
2834: PetscInt nz = a->i[m], row, *jj, mr, col;
2836: PetscFunctionBegin;
2837: *nn = n;
2838: if (!ia) PetscFunctionReturn(PETSC_SUCCESS);
2839: PetscCheck(!symmetric, PETSC_COMM_SELF, PETSC_ERR_SUP, "Not for BAIJ matrices");
2840: PetscCall(PetscCalloc1(n, &collengths));
2841: PetscCall(PetscMalloc1(n + 1, &cia));
2842: PetscCall(PetscMalloc1(nz, &cja));
2843: jj = a->j;
2844: for (i = 0; i < nz; i++) collengths[jj[i]]++;
2845: cia[0] = oshift;
2846: for (i = 0; i < n; i++) cia[i + 1] = cia[i] + collengths[i];
2847: PetscCall(PetscArrayzero(collengths, n));
2848: jj = a->j;
2849: for (row = 0; row < m; row++) {
2850: mr = a->i[row + 1] - a->i[row];
2851: for (i = 0; i < mr; i++) {
2852: col = *jj++;
2854: cja[cia[col] + collengths[col]++ - oshift] = row + oshift;
2855: }
2856: }
2857: PetscCall(PetscFree(collengths));
2858: *ia = cia;
2859: *ja = cja;
2860: PetscFunctionReturn(PETSC_SUCCESS);
2861: }
2863: static PetscErrorCode MatRestoreColumnIJ_SeqBAIJ(Mat A, PetscInt oshift, PetscBool symmetric, PetscBool inodecompressed, PetscInt *n, const PetscInt *ia[], const PetscInt *ja[], PetscBool *done)
2864: {
2865: PetscFunctionBegin;
2866: if (!ia) PetscFunctionReturn(PETSC_SUCCESS);
2867: PetscCall(PetscFree(*ia));
2868: PetscCall(PetscFree(*ja));
2869: PetscFunctionReturn(PETSC_SUCCESS);
2870: }
2872: /*
2873: MatGetColumnIJ_SeqBAIJ_Color() and MatRestoreColumnIJ_SeqBAIJ_Color() are customized from
2874: MatGetColumnIJ_SeqBAIJ() and MatRestoreColumnIJ_SeqBAIJ() by adding an output
2875: spidx[], index of a->a, to be used in MatTransposeColoringCreate() and MatFDColoringCreate()
2876: */
2877: PetscErrorCode MatGetColumnIJ_SeqBAIJ_Color(Mat A, PetscInt oshift, PetscBool symmetric, PetscBool inodecompressed, PetscInt *nn, const PetscInt *ia[], const PetscInt *ja[], PetscInt *spidx[], PetscBool *done)
2878: {
2879: Mat_SeqBAIJ *a = (Mat_SeqBAIJ *)A->data;
2880: PetscInt i, *collengths, *cia, *cja, n = a->nbs, m = a->mbs;
2881: PetscInt nz = a->i[m], row, *jj, mr, col;
2882: PetscInt *cspidx;
2884: PetscFunctionBegin;
2885: *nn = n;
2886: if (!ia) PetscFunctionReturn(PETSC_SUCCESS);
2888: PetscCall(PetscCalloc1(n, &collengths));
2889: PetscCall(PetscMalloc1(n + 1, &cia));
2890: PetscCall(PetscMalloc1(nz, &cja));
2891: PetscCall(PetscMalloc1(nz, &cspidx));
2892: jj = a->j;
2893: for (i = 0; i < nz; i++) collengths[jj[i]]++;
2894: cia[0] = oshift;
2895: for (i = 0; i < n; i++) cia[i + 1] = cia[i] + collengths[i];
2896: PetscCall(PetscArrayzero(collengths, n));
2897: jj = a->j;
2898: for (row = 0; row < m; row++) {
2899: mr = a->i[row + 1] - a->i[row];
2900: for (i = 0; i < mr; i++) {
2901: col = *jj++;
2902: cspidx[cia[col] + collengths[col] - oshift] = a->i[row] + i; /* index of a->j */
2903: cja[cia[col] + collengths[col]++ - oshift] = row + oshift;
2904: }
2905: }
2906: PetscCall(PetscFree(collengths));
2907: *ia = cia;
2908: *ja = cja;
2909: *spidx = cspidx;
2910: PetscFunctionReturn(PETSC_SUCCESS);
2911: }
2913: PetscErrorCode MatRestoreColumnIJ_SeqBAIJ_Color(Mat A, PetscInt oshift, PetscBool symmetric, PetscBool inodecompressed, PetscInt *n, const PetscInt *ia[], const PetscInt *ja[], PetscInt *spidx[], PetscBool *done)
2914: {
2915: PetscFunctionBegin;
2916: PetscCall(MatRestoreColumnIJ_SeqBAIJ(A, oshift, symmetric, inodecompressed, n, ia, ja, done));
2917: PetscCall(PetscFree(*spidx));
2918: PetscFunctionReturn(PETSC_SUCCESS);
2919: }
2921: PetscErrorCode MatShift_SeqBAIJ(Mat Y, PetscScalar a)
2922: {
2923: Mat_SeqBAIJ *aij = (Mat_SeqBAIJ *)Y->data;
2925: PetscFunctionBegin;
2926: if (!Y->preallocated || !aij->nz) PetscCall(MatSeqBAIJSetPreallocation(Y, Y->rmap->bs, 1, NULL));
2927: PetscCall(MatShift_Basic(Y, a));
2928: PetscFunctionReturn(PETSC_SUCCESS);
2929: }
2931: static struct _MatOps MatOps_Values = {MatSetValues_SeqBAIJ,
2932: MatGetRow_SeqBAIJ,
2933: MatRestoreRow_SeqBAIJ,
2934: MatMult_SeqBAIJ_N,
2935: /* 4*/ MatMultAdd_SeqBAIJ_N,
2936: MatMultTranspose_SeqBAIJ,
2937: MatMultTransposeAdd_SeqBAIJ,
2938: NULL,
2939: NULL,
2940: NULL,
2941: /* 10*/ NULL,
2942: MatLUFactor_SeqBAIJ,
2943: NULL,
2944: NULL,
2945: MatTranspose_SeqBAIJ,
2946: /* 15*/ MatGetInfo_SeqBAIJ,
2947: MatEqual_SeqBAIJ,
2948: MatGetDiagonal_SeqBAIJ,
2949: MatDiagonalScale_SeqBAIJ,
2950: MatNorm_SeqBAIJ,
2951: /* 20*/ NULL,
2952: MatAssemblyEnd_SeqBAIJ,
2953: MatSetOption_SeqBAIJ,
2954: MatZeroEntries_SeqBAIJ,
2955: /* 24*/ MatZeroRows_SeqBAIJ,
2956: NULL,
2957: NULL,
2958: NULL,
2959: NULL,
2960: /* 29*/ MatSetUp_Seq_Hash,
2961: NULL,
2962: NULL,
2963: NULL,
2964: NULL,
2965: /* 34*/ MatDuplicate_SeqBAIJ,
2966: NULL,
2967: NULL,
2968: MatILUFactor_SeqBAIJ,
2969: NULL,
2970: /* 39*/ MatAXPY_SeqBAIJ,
2971: MatCreateSubMatrices_SeqBAIJ,
2972: MatIncreaseOverlap_SeqBAIJ,
2973: MatGetValues_SeqBAIJ,
2974: MatCopy_SeqBAIJ,
2975: /* 44*/ NULL,
2976: MatScale_SeqBAIJ,
2977: MatShift_SeqBAIJ,
2978: NULL,
2979: MatZeroRowsColumns_SeqBAIJ,
2980: /* 49*/ NULL,
2981: MatGetRowIJ_SeqBAIJ,
2982: MatRestoreRowIJ_SeqBAIJ,
2983: MatGetColumnIJ_SeqBAIJ,
2984: MatRestoreColumnIJ_SeqBAIJ,
2985: /* 54*/ MatFDColoringCreate_SeqXAIJ,
2986: NULL,
2987: NULL,
2988: NULL,
2989: MatSetValuesBlocked_SeqBAIJ,
2990: /* 59*/ MatCreateSubMatrix_SeqBAIJ,
2991: MatDestroy_SeqBAIJ,
2992: MatView_SeqBAIJ,
2993: NULL,
2994: NULL,
2995: /* 64*/ NULL,
2996: NULL,
2997: NULL,
2998: NULL,
2999: NULL,
3000: /* 69*/ MatGetRowMaxAbs_SeqBAIJ,
3001: NULL,
3002: MatConvert_Basic,
3003: NULL,
3004: NULL,
3005: /* 74*/ NULL,
3006: MatFDColoringApply_BAIJ,
3007: NULL,
3008: NULL,
3009: NULL,
3010: /* 79*/ NULL,
3011: NULL,
3012: NULL,
3013: NULL,
3014: MatLoad_SeqBAIJ,
3015: /* 84*/ NULL,
3016: NULL,
3017: NULL,
3018: NULL,
3019: NULL,
3020: /* 89*/ NULL,
3021: NULL,
3022: NULL,
3023: NULL,
3024: NULL,
3025: /* 94*/ NULL,
3026: NULL,
3027: NULL,
3028: NULL,
3029: NULL,
3030: /* 99*/ NULL,
3031: NULL,
3032: NULL,
3033: MatConjugate_SeqBAIJ,
3034: NULL,
3035: /*104*/ NULL,
3036: MatRealPart_SeqBAIJ,
3037: MatImaginaryPart_SeqBAIJ,
3038: NULL,
3039: NULL,
3040: /*109*/ NULL,
3041: NULL,
3042: NULL,
3043: NULL,
3044: MatMissingDiagonal_SeqBAIJ,
3045: /*114*/ NULL,
3046: NULL,
3047: NULL,
3048: NULL,
3049: NULL,
3050: /*119*/ NULL,
3051: NULL,
3052: MatMultHermitianTranspose_SeqBAIJ,
3053: MatMultHermitianTransposeAdd_SeqBAIJ,
3054: NULL,
3055: /*124*/ NULL,
3056: MatGetColumnReductions_SeqBAIJ,
3057: MatInvertBlockDiagonal_SeqBAIJ,
3058: NULL,
3059: NULL,
3060: /*129*/ NULL,
3061: NULL,
3062: NULL,
3063: NULL,
3064: NULL,
3065: /*134*/ NULL,
3066: NULL,
3067: NULL,
3068: NULL,
3069: NULL,
3070: /*139*/ MatSetBlockSizes_Default,
3071: NULL,
3072: NULL,
3073: MatFDColoringSetUp_SeqXAIJ,
3074: NULL,
3075: /*144*/ MatCreateMPIMatConcatenateSeqMat_SeqBAIJ,
3076: MatDestroySubMatrices_SeqBAIJ,
3077: NULL,
3078: NULL,
3079: NULL,
3080: NULL,
3081: /*150*/ NULL,
3082: NULL};
3084: static PetscErrorCode MatStoreValues_SeqBAIJ(Mat mat)
3085: {
3086: Mat_SeqBAIJ *aij = (Mat_SeqBAIJ *)mat->data;
3087: PetscInt nz = aij->i[aij->mbs] * aij->bs2;
3089: PetscFunctionBegin;
3090: PetscCheck(aij->nonew == 1, PETSC_COMM_SELF, PETSC_ERR_ORDER, "Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first");
3092: /* allocate space for values if not already there */
3093: if (!aij->saved_values) PetscCall(PetscMalloc1(nz + 1, &aij->saved_values));
3095: /* copy values over */
3096: PetscCall(PetscArraycpy(aij->saved_values, aij->a, nz));
3097: PetscFunctionReturn(PETSC_SUCCESS);
3098: }
3100: static PetscErrorCode MatRetrieveValues_SeqBAIJ(Mat mat)
3101: {
3102: Mat_SeqBAIJ *aij = (Mat_SeqBAIJ *)mat->data;
3103: PetscInt nz = aij->i[aij->mbs] * aij->bs2;
3105: PetscFunctionBegin;
3106: PetscCheck(aij->nonew == 1, PETSC_COMM_SELF, PETSC_ERR_ORDER, "Must call MatSetOption(A,MAT_NEW_NONZERO_LOCATIONS,PETSC_FALSE);first");
3107: PetscCheck(aij->saved_values, PETSC_COMM_SELF, PETSC_ERR_ORDER, "Must call MatStoreValues(A);first");
3109: /* copy values over */
3110: PetscCall(PetscArraycpy(aij->a, aij->saved_values, nz));
3111: PetscFunctionReturn(PETSC_SUCCESS);
3112: }
3114: PETSC_INTERN PetscErrorCode MatConvert_SeqBAIJ_SeqAIJ(Mat, MatType, MatReuse, Mat *);
3115: PETSC_INTERN PetscErrorCode MatConvert_SeqBAIJ_SeqSBAIJ(Mat, MatType, MatReuse, Mat *);
3117: PetscErrorCode MatSeqBAIJSetPreallocation_SeqBAIJ(Mat B, PetscInt bs, PetscInt nz, PetscInt *nnz)
3118: {
3119: Mat_SeqBAIJ *b = (Mat_SeqBAIJ *)B->data;
3120: PetscInt i, mbs, nbs, bs2;
3121: PetscBool flg = PETSC_FALSE, skipallocation = PETSC_FALSE, realalloc = PETSC_FALSE;
3123: PetscFunctionBegin;
3124: if (B->hash_active) {
3125: PetscInt bs;
3126: PetscCall(PetscMemcpy(&B->ops, &b->cops, sizeof(*(B->ops))));
3127: PetscCall(PetscHMapIJVDestroy(&b->ht));
3128: PetscCall(MatGetBlockSize(B, &bs));
3129: if (bs > 1) PetscCall(PetscHSetIJDestroy(&b->bht));
3130: PetscCall(PetscFree(b->dnz));
3131: PetscCall(PetscFree(b->bdnz));
3132: B->hash_active = PETSC_FALSE;
3133: }
3134: if (nz >= 0 || nnz) realalloc = PETSC_TRUE;
3135: if (nz == MAT_SKIP_ALLOCATION) {
3136: skipallocation = PETSC_TRUE;
3137: nz = 0;
3138: }
3140: PetscCall(MatSetBlockSize(B, PetscAbs(bs)));
3141: PetscCall(PetscLayoutSetUp(B->rmap));
3142: PetscCall(PetscLayoutSetUp(B->cmap));
3143: PetscCall(PetscLayoutGetBlockSize(B->rmap, &bs));
3145: B->preallocated = PETSC_TRUE;
3147: mbs = B->rmap->n / bs;
3148: nbs = B->cmap->n / bs;
3149: bs2 = bs * bs;
3151: PetscCheck(mbs * bs == B->rmap->n && nbs * bs == B->cmap->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Number rows %" PetscInt_FMT ", cols %" PetscInt_FMT " must be divisible by blocksize %" PetscInt_FMT, B->rmap->N, B->cmap->n, bs);
3153: if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 5;
3154: PetscCheck(nz >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "nz cannot be less than 0: value %" PetscInt_FMT, nz);
3155: if (nnz) {
3156: for (i = 0; i < mbs; i++) {
3157: PetscCheck(nnz[i] >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "nnz cannot be less than 0: local row %" PetscInt_FMT " value %" PetscInt_FMT, i, nnz[i]);
3158: PetscCheck(nnz[i] <= nbs, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "nnz cannot be greater than block row length: local row %" PetscInt_FMT " value %" PetscInt_FMT " rowlength %" PetscInt_FMT, i, nnz[i], nbs);
3159: }
3160: }
3162: PetscOptionsBegin(PetscObjectComm((PetscObject)B), NULL, "Optimize options for SEQBAIJ matrix 2 ", "Mat");
3163: PetscCall(PetscOptionsBool("-mat_no_unroll", "Do not optimize for block size (slow)", NULL, flg, &flg, NULL));
3164: PetscOptionsEnd();
3166: if (!flg) {
3167: switch (bs) {
3168: case 1:
3169: B->ops->mult = MatMult_SeqBAIJ_1;
3170: B->ops->multadd = MatMultAdd_SeqBAIJ_1;
3171: break;
3172: case 2:
3173: B->ops->mult = MatMult_SeqBAIJ_2;
3174: B->ops->multadd = MatMultAdd_SeqBAIJ_2;
3175: break;
3176: case 3:
3177: B->ops->mult = MatMult_SeqBAIJ_3;
3178: B->ops->multadd = MatMultAdd_SeqBAIJ_3;
3179: break;
3180: case 4:
3181: B->ops->mult = MatMult_SeqBAIJ_4;
3182: B->ops->multadd = MatMultAdd_SeqBAIJ_4;
3183: break;
3184: case 5:
3185: B->ops->mult = MatMult_SeqBAIJ_5;
3186: B->ops->multadd = MatMultAdd_SeqBAIJ_5;
3187: break;
3188: case 6:
3189: B->ops->mult = MatMult_SeqBAIJ_6;
3190: B->ops->multadd = MatMultAdd_SeqBAIJ_6;
3191: break;
3192: case 7:
3193: B->ops->mult = MatMult_SeqBAIJ_7;
3194: B->ops->multadd = MatMultAdd_SeqBAIJ_7;
3195: break;
3196: case 9: {
3197: PetscInt version = 1;
3198: PetscCall(PetscOptionsGetInt(NULL, ((PetscObject)B)->prefix, "-mat_baij_mult_version", &version, NULL));
3199: switch (version) {
3200: #if defined(PETSC_HAVE_IMMINTRIN_H) && defined(__AVX2__) && defined(__FMA__) && defined(PETSC_USE_REAL_DOUBLE) && !defined(PETSC_USE_COMPLEX) && !defined(PETSC_USE_64BIT_INDICES)
3201: case 1:
3202: B->ops->mult = MatMult_SeqBAIJ_9_AVX2;
3203: B->ops->multadd = MatMultAdd_SeqBAIJ_9_AVX2;
3204: PetscCall(PetscInfo((PetscObject)B, "Using AVX2 for MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", bs));
3205: break;
3206: #endif
3207: default:
3208: B->ops->mult = MatMult_SeqBAIJ_N;
3209: B->ops->multadd = MatMultAdd_SeqBAIJ_N;
3210: PetscCall(PetscInfo((PetscObject)B, "Using BLAS for MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", bs));
3211: break;
3212: }
3213: break;
3214: }
3215: case 11:
3216: B->ops->mult = MatMult_SeqBAIJ_11;
3217: B->ops->multadd = MatMultAdd_SeqBAIJ_11;
3218: break;
3219: case 12: {
3220: PetscInt version = 1;
3221: PetscCall(PetscOptionsGetInt(NULL, ((PetscObject)B)->prefix, "-mat_baij_mult_version", &version, NULL));
3222: switch (version) {
3223: case 1:
3224: B->ops->mult = MatMult_SeqBAIJ_12_ver1;
3225: B->ops->multadd = MatMultAdd_SeqBAIJ_12_ver1;
3226: PetscCall(PetscInfo((PetscObject)B, "Using version %" PetscInt_FMT " of MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", version, bs));
3227: break;
3228: case 2:
3229: B->ops->mult = MatMult_SeqBAIJ_12_ver2;
3230: B->ops->multadd = MatMultAdd_SeqBAIJ_12_ver2;
3231: PetscCall(PetscInfo((PetscObject)B, "Using version %" PetscInt_FMT " of MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", version, bs));
3232: break;
3233: #if defined(PETSC_HAVE_IMMINTRIN_H) && defined(__AVX2__) && defined(__FMA__) && defined(PETSC_USE_REAL_DOUBLE) && !defined(PETSC_USE_COMPLEX) && !defined(PETSC_USE_64BIT_INDICES)
3234: case 3:
3235: B->ops->mult = MatMult_SeqBAIJ_12_AVX2;
3236: B->ops->multadd = MatMultAdd_SeqBAIJ_12_ver1;
3237: PetscCall(PetscInfo((PetscObject)B, "Using AVX2 for MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", bs));
3238: break;
3239: #endif
3240: default:
3241: B->ops->mult = MatMult_SeqBAIJ_N;
3242: B->ops->multadd = MatMultAdd_SeqBAIJ_N;
3243: PetscCall(PetscInfo((PetscObject)B, "Using BLAS for MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", bs));
3244: break;
3245: }
3246: break;
3247: }
3248: case 15: {
3249: PetscInt version = 1;
3250: PetscCall(PetscOptionsGetInt(NULL, ((PetscObject)B)->prefix, "-mat_baij_mult_version", &version, NULL));
3251: switch (version) {
3252: case 1:
3253: B->ops->mult = MatMult_SeqBAIJ_15_ver1;
3254: PetscCall(PetscInfo((PetscObject)B, "Using version %" PetscInt_FMT " of MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", version, bs));
3255: break;
3256: case 2:
3257: B->ops->mult = MatMult_SeqBAIJ_15_ver2;
3258: PetscCall(PetscInfo((PetscObject)B, "Using version %" PetscInt_FMT " of MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", version, bs));
3259: break;
3260: case 3:
3261: B->ops->mult = MatMult_SeqBAIJ_15_ver3;
3262: PetscCall(PetscInfo((PetscObject)B, "Using version %" PetscInt_FMT " of MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", version, bs));
3263: break;
3264: case 4:
3265: B->ops->mult = MatMult_SeqBAIJ_15_ver4;
3266: PetscCall(PetscInfo((PetscObject)B, "Using version %" PetscInt_FMT " of MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", version, bs));
3267: break;
3268: default:
3269: B->ops->mult = MatMult_SeqBAIJ_N;
3270: PetscCall(PetscInfo((PetscObject)B, "Using BLAS for MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", bs));
3271: break;
3272: }
3273: B->ops->multadd = MatMultAdd_SeqBAIJ_N;
3274: break;
3275: }
3276: default:
3277: B->ops->mult = MatMult_SeqBAIJ_N;
3278: B->ops->multadd = MatMultAdd_SeqBAIJ_N;
3279: PetscCall(PetscInfo((PetscObject)B, "Using BLAS for MatMult for BAIJ for blocksize %" PetscInt_FMT "\n", bs));
3280: break;
3281: }
3282: }
3283: B->ops->sor = MatSOR_SeqBAIJ;
3284: b->mbs = mbs;
3285: b->nbs = nbs;
3286: if (!skipallocation) {
3287: if (!b->imax) {
3288: PetscCall(PetscMalloc2(mbs, &b->imax, mbs, &b->ilen));
3290: b->free_imax_ilen = PETSC_TRUE;
3291: }
3292: /* b->ilen will count nonzeros in each block row so far. */
3293: for (i = 0; i < mbs; i++) b->ilen[i] = 0;
3294: if (!nnz) {
3295: if (nz == PETSC_DEFAULT || nz == PETSC_DECIDE) nz = 5;
3296: else if (nz < 0) nz = 1;
3297: nz = PetscMin(nz, nbs);
3298: for (i = 0; i < mbs; i++) b->imax[i] = nz;
3299: PetscCall(PetscIntMultError(nz, mbs, &nz));
3300: } else {
3301: PetscInt64 nz64 = 0;
3302: for (i = 0; i < mbs; i++) {
3303: b->imax[i] = nnz[i];
3304: nz64 += nnz[i];
3305: }
3306: PetscCall(PetscIntCast(nz64, &nz));
3307: }
3309: /* allocate the matrix space */
3310: PetscCall(MatSeqXAIJFreeAIJ(B, &b->a, &b->j, &b->i));
3311: if (B->structure_only) {
3312: PetscCall(PetscMalloc1(nz, &b->j));
3313: PetscCall(PetscMalloc1(B->rmap->N + 1, &b->i));
3314: } else {
3315: PetscInt nzbs2 = 0;
3316: PetscCall(PetscIntMultError(nz, bs2, &nzbs2));
3317: PetscCall(PetscMalloc3(nzbs2, &b->a, nz, &b->j, B->rmap->N + 1, &b->i));
3318: PetscCall(PetscArrayzero(b->a, nz * bs2));
3319: }
3320: PetscCall(PetscArrayzero(b->j, nz));
3322: if (B->structure_only) {
3323: b->singlemalloc = PETSC_FALSE;
3324: b->free_a = PETSC_FALSE;
3325: } else {
3326: b->singlemalloc = PETSC_TRUE;
3327: b->free_a = PETSC_TRUE;
3328: }
3329: b->free_ij = PETSC_TRUE;
3331: b->i[0] = 0;
3332: for (i = 1; i < mbs + 1; i++) b->i[i] = b->i[i - 1] + b->imax[i - 1];
3334: } else {
3335: b->free_a = PETSC_FALSE;
3336: b->free_ij = PETSC_FALSE;
3337: }
3339: b->bs2 = bs2;
3340: b->mbs = mbs;
3341: b->nz = 0;
3342: b->maxnz = nz;
3343: B->info.nz_unneeded = (PetscReal)b->maxnz * bs2;
3344: B->was_assembled = PETSC_FALSE;
3345: B->assembled = PETSC_FALSE;
3346: if (realalloc) PetscCall(MatSetOption(B, MAT_NEW_NONZERO_ALLOCATION_ERR, PETSC_TRUE));
3347: PetscFunctionReturn(PETSC_SUCCESS);
3348: }
3350: PetscErrorCode MatSeqBAIJSetPreallocationCSR_SeqBAIJ(Mat B, PetscInt bs, const PetscInt ii[], const PetscInt jj[], const PetscScalar V[])
3351: {
3352: PetscInt i, m, nz, nz_max = 0, *nnz;
3353: PetscScalar *values = NULL;
3354: PetscBool roworiented = ((Mat_SeqBAIJ *)B->data)->roworiented;
3356: PetscFunctionBegin;
3357: PetscCheck(bs >= 1, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Invalid block size specified, must be positive but it is %" PetscInt_FMT, bs);
3358: PetscCall(PetscLayoutSetBlockSize(B->rmap, bs));
3359: PetscCall(PetscLayoutSetBlockSize(B->cmap, bs));
3360: PetscCall(PetscLayoutSetUp(B->rmap));
3361: PetscCall(PetscLayoutSetUp(B->cmap));
3362: PetscCall(PetscLayoutGetBlockSize(B->rmap, &bs));
3363: m = B->rmap->n / bs;
3365: PetscCheck(ii[0] == 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "ii[0] must be 0 but it is %" PetscInt_FMT, ii[0]);
3366: PetscCall(PetscMalloc1(m + 1, &nnz));
3367: for (i = 0; i < m; i++) {
3368: nz = ii[i + 1] - ii[i];
3369: PetscCheck(nz >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Local row %" PetscInt_FMT " has a negative number of columns %" PetscInt_FMT, i, nz);
3370: nz_max = PetscMax(nz_max, nz);
3371: nnz[i] = nz;
3372: }
3373: PetscCall(MatSeqBAIJSetPreallocation(B, bs, 0, nnz));
3374: PetscCall(PetscFree(nnz));
3376: values = (PetscScalar *)V;
3377: if (!values) PetscCall(PetscCalloc1(bs * bs * (nz_max + 1), &values));
3378: for (i = 0; i < m; i++) {
3379: PetscInt ncols = ii[i + 1] - ii[i];
3380: const PetscInt *icols = jj + ii[i];
3381: if (bs == 1 || !roworiented) {
3382: const PetscScalar *svals = values + (V ? (bs * bs * ii[i]) : 0);
3383: PetscCall(MatSetValuesBlocked_SeqBAIJ(B, 1, &i, ncols, icols, svals, INSERT_VALUES));
3384: } else {
3385: PetscInt j;
3386: for (j = 0; j < ncols; j++) {
3387: const PetscScalar *svals = values + (V ? (bs * bs * (ii[i] + j)) : 0);
3388: PetscCall(MatSetValuesBlocked_SeqBAIJ(B, 1, &i, 1, &icols[j], svals, INSERT_VALUES));
3389: }
3390: }
3391: }
3392: if (!V) PetscCall(PetscFree(values));
3393: PetscCall(MatAssemblyBegin(B, MAT_FINAL_ASSEMBLY));
3394: PetscCall(MatAssemblyEnd(B, MAT_FINAL_ASSEMBLY));
3395: PetscCall(MatSetOption(B, MAT_NEW_NONZERO_LOCATION_ERR, PETSC_TRUE));
3396: PetscFunctionReturn(PETSC_SUCCESS);
3397: }
3399: /*@C
3400: MatSeqBAIJGetArray - gives read/write access to the array where the data for a `MATSEQBAIJ` matrix is stored
3402: Not Collective
3404: Input Parameter:
3405: . mat - a `MATSEQBAIJ` matrix
3407: Output Parameter:
3408: . array - pointer to the data
3410: Level: intermediate
3412: .seealso: [](ch_matrices), `Mat`, `MATSEQBAIJ`, `MatSeqBAIJRestoreArray()`, `MatSeqAIJGetArray()`, `MatSeqAIJRestoreArray()`
3413: @*/
3414: PetscErrorCode MatSeqBAIJGetArray(Mat A, PetscScalar **array)
3415: {
3416: PetscFunctionBegin;
3417: PetscUseMethod(A, "MatSeqBAIJGetArray_C", (Mat, PetscScalar **), (A, array));
3418: PetscFunctionReturn(PETSC_SUCCESS);
3419: }
3421: /*@C
3422: MatSeqBAIJRestoreArray - returns access to the array where the data for a `MATSEQBAIJ` matrix is stored obtained by `MatSeqBAIJGetArray()`
3424: Not Collective
3426: Input Parameters:
3427: + mat - a `MATSEQBAIJ` matrix
3428: - array - pointer to the data
3430: Level: intermediate
3432: .seealso: [](ch_matrices), `Mat`, `MatSeqBAIJGetArray()`, `MatSeqAIJGetArray()`, `MatSeqAIJRestoreArray()`
3433: @*/
3434: PetscErrorCode MatSeqBAIJRestoreArray(Mat A, PetscScalar **array)
3435: {
3436: PetscFunctionBegin;
3437: PetscUseMethod(A, "MatSeqBAIJRestoreArray_C", (Mat, PetscScalar **), (A, array));
3438: PetscFunctionReturn(PETSC_SUCCESS);
3439: }
3441: /*MC
3442: MATSEQBAIJ - MATSEQBAIJ = "seqbaij" - A matrix type to be used for sequential block sparse matrices, based on
3443: block sparse compressed row format.
3445: Options Database Keys:
3446: + -mat_type seqbaij - sets the matrix type to `MATSEQBAIJ` during a call to `MatSetFromOptions()`
3447: - -mat_baij_mult_version version - indicate the version of the matrix-vector product to use (0 often indicates using BLAS)
3449: Level: beginner
3451: Notes:
3452: `MatSetOptions`(,`MAT_STRUCTURE_ONLY`,`PETSC_TRUE`) may be called for this matrix type. In this no
3453: space is allocated for the nonzero entries and any entries passed with `MatSetValues()` are ignored
3455: Run with `-info` to see what version of the matrix-vector product is being used
3457: .seealso: [](ch_matrices), `Mat`, `MatCreateSeqBAIJ()`
3458: M*/
3460: PETSC_INTERN PetscErrorCode MatConvert_SeqBAIJ_SeqBSTRM(Mat, MatType, MatReuse, Mat *);
3462: PETSC_EXTERN PetscErrorCode MatCreate_SeqBAIJ(Mat B)
3463: {
3464: PetscMPIInt size;
3465: Mat_SeqBAIJ *b;
3467: PetscFunctionBegin;
3468: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)B), &size));
3469: PetscCheck(size == 1, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "Comm must be of size 1");
3471: PetscCall(PetscNew(&b));
3472: B->data = (void *)b;
3473: PetscCall(PetscMemcpy(B->ops, &MatOps_Values, sizeof(struct _MatOps)));
3475: b->row = NULL;
3476: b->col = NULL;
3477: b->icol = NULL;
3478: b->reallocs = 0;
3479: b->saved_values = NULL;
3481: b->roworiented = PETSC_TRUE;
3482: b->nonew = 0;
3483: b->diag = NULL;
3484: B->spptr = NULL;
3485: B->info.nz_unneeded = (PetscReal)b->maxnz * b->bs2;
3486: b->keepnonzeropattern = PETSC_FALSE;
3488: PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatSeqBAIJGetArray_C", MatSeqBAIJGetArray_SeqBAIJ));
3489: PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatSeqBAIJRestoreArray_C", MatSeqBAIJRestoreArray_SeqBAIJ));
3490: PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatStoreValues_C", MatStoreValues_SeqBAIJ));
3491: PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatRetrieveValues_C", MatRetrieveValues_SeqBAIJ));
3492: PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatSeqBAIJSetColumnIndices_C", MatSeqBAIJSetColumnIndices_SeqBAIJ));
3493: PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatConvert_seqbaij_seqaij_C", MatConvert_SeqBAIJ_SeqAIJ));
3494: PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatConvert_seqbaij_seqsbaij_C", MatConvert_SeqBAIJ_SeqSBAIJ));
3495: PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatSeqBAIJSetPreallocation_C", MatSeqBAIJSetPreallocation_SeqBAIJ));
3496: PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatSeqBAIJSetPreallocationCSR_C", MatSeqBAIJSetPreallocationCSR_SeqBAIJ));
3497: PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatIsTranspose_C", MatIsTranspose_SeqBAIJ));
3498: #if defined(PETSC_HAVE_HYPRE)
3499: PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatConvert_seqbaij_hypre_C", MatConvert_AIJ_HYPRE));
3500: #endif
3501: PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatConvert_seqbaij_is_C", MatConvert_XAIJ_IS));
3502: PetscCall(PetscObjectChangeTypeName((PetscObject)B, MATSEQBAIJ));
3503: PetscFunctionReturn(PETSC_SUCCESS);
3504: }
3506: PetscErrorCode MatDuplicateNoCreate_SeqBAIJ(Mat C, Mat A, MatDuplicateOption cpvalues, PetscBool mallocmatspace)
3507: {
3508: Mat_SeqBAIJ *c = (Mat_SeqBAIJ *)C->data, *a = (Mat_SeqBAIJ *)A->data;
3509: PetscInt i, mbs = a->mbs, nz = a->nz, bs2 = a->bs2;
3511: PetscFunctionBegin;
3512: PetscCheck(A->assembled, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONGSTATE, "Cannot duplicate unassembled matrix");
3513: PetscCheck(a->i[mbs] == nz, PETSC_COMM_SELF, PETSC_ERR_PLIB, "Corrupt matrix");
3515: if (cpvalues == MAT_SHARE_NONZERO_PATTERN) {
3516: c->imax = a->imax;
3517: c->ilen = a->ilen;
3518: c->free_imax_ilen = PETSC_FALSE;
3519: } else {
3520: PetscCall(PetscMalloc2(mbs, &c->imax, mbs, &c->ilen));
3521: for (i = 0; i < mbs; i++) {
3522: c->imax[i] = a->imax[i];
3523: c->ilen[i] = a->ilen[i];
3524: }
3525: c->free_imax_ilen = PETSC_TRUE;
3526: }
3528: /* allocate the matrix space */
3529: if (mallocmatspace) {
3530: if (cpvalues == MAT_SHARE_NONZERO_PATTERN) {
3531: PetscCall(PetscCalloc1(bs2 * nz, &c->a));
3533: c->i = a->i;
3534: c->j = a->j;
3535: c->singlemalloc = PETSC_FALSE;
3536: c->free_a = PETSC_TRUE;
3537: c->free_ij = PETSC_FALSE;
3538: c->parent = A;
3539: C->preallocated = PETSC_TRUE;
3540: C->assembled = PETSC_TRUE;
3542: PetscCall(PetscObjectReference((PetscObject)A));
3543: PetscCall(MatSetOption(A, MAT_NEW_NONZERO_LOCATION_ERR, PETSC_TRUE));
3544: PetscCall(MatSetOption(C, MAT_NEW_NONZERO_LOCATION_ERR, PETSC_TRUE));
3545: } else {
3546: PetscCall(PetscMalloc3(bs2 * nz, &c->a, nz, &c->j, mbs + 1, &c->i));
3548: c->singlemalloc = PETSC_TRUE;
3549: c->free_a = PETSC_TRUE;
3550: c->free_ij = PETSC_TRUE;
3552: PetscCall(PetscArraycpy(c->i, a->i, mbs + 1));
3553: if (mbs > 0) {
3554: PetscCall(PetscArraycpy(c->j, a->j, nz));
3555: if (cpvalues == MAT_COPY_VALUES) {
3556: PetscCall(PetscArraycpy(c->a, a->a, bs2 * nz));
3557: } else {
3558: PetscCall(PetscArrayzero(c->a, bs2 * nz));
3559: }
3560: }
3561: C->preallocated = PETSC_TRUE;
3562: C->assembled = PETSC_TRUE;
3563: }
3564: }
3566: c->roworiented = a->roworiented;
3567: c->nonew = a->nonew;
3569: PetscCall(PetscLayoutReference(A->rmap, &C->rmap));
3570: PetscCall(PetscLayoutReference(A->cmap, &C->cmap));
3572: c->bs2 = a->bs2;
3573: c->mbs = a->mbs;
3574: c->nbs = a->nbs;
3576: if (a->diag) {
3577: if (cpvalues == MAT_SHARE_NONZERO_PATTERN) {
3578: c->diag = a->diag;
3579: c->free_diag = PETSC_FALSE;
3580: } else {
3581: PetscCall(PetscMalloc1(mbs + 1, &c->diag));
3582: for (i = 0; i < mbs; i++) c->diag[i] = a->diag[i];
3583: c->free_diag = PETSC_TRUE;
3584: }
3585: } else c->diag = NULL;
3587: c->nz = a->nz;
3588: c->maxnz = a->nz; /* Since we allocate exactly the right amount */
3589: c->solve_work = NULL;
3590: c->mult_work = NULL;
3591: c->sor_workt = NULL;
3592: c->sor_work = NULL;
3594: c->compressedrow.use = a->compressedrow.use;
3595: c->compressedrow.nrows = a->compressedrow.nrows;
3596: if (a->compressedrow.use) {
3597: i = a->compressedrow.nrows;
3598: PetscCall(PetscMalloc2(i + 1, &c->compressedrow.i, i + 1, &c->compressedrow.rindex));
3599: PetscCall(PetscArraycpy(c->compressedrow.i, a->compressedrow.i, i + 1));
3600: PetscCall(PetscArraycpy(c->compressedrow.rindex, a->compressedrow.rindex, i));
3601: } else {
3602: c->compressedrow.use = PETSC_FALSE;
3603: c->compressedrow.i = NULL;
3604: c->compressedrow.rindex = NULL;
3605: }
3606: C->nonzerostate = A->nonzerostate;
3608: PetscCall(PetscFunctionListDuplicate(((PetscObject)A)->qlist, &((PetscObject)C)->qlist));
3609: PetscFunctionReturn(PETSC_SUCCESS);
3610: }
3612: PetscErrorCode MatDuplicate_SeqBAIJ(Mat A, MatDuplicateOption cpvalues, Mat *B)
3613: {
3614: PetscFunctionBegin;
3615: PetscCall(MatCreate(PetscObjectComm((PetscObject)A), B));
3616: PetscCall(MatSetSizes(*B, A->rmap->N, A->cmap->n, A->rmap->N, A->cmap->n));
3617: PetscCall(MatSetType(*B, MATSEQBAIJ));
3618: PetscCall(MatDuplicateNoCreate_SeqBAIJ(*B, A, cpvalues, PETSC_TRUE));
3619: PetscFunctionReturn(PETSC_SUCCESS);
3620: }
3622: /* Used for both SeqBAIJ and SeqSBAIJ matrices */
3623: PetscErrorCode MatLoad_SeqBAIJ_Binary(Mat mat, PetscViewer viewer)
3624: {
3625: PetscInt header[4], M, N, nz, bs, m, n, mbs, nbs, rows, cols, sum, i, j, k;
3626: PetscInt *rowidxs, *colidxs;
3627: PetscScalar *matvals;
3629: PetscFunctionBegin;
3630: PetscCall(PetscViewerSetUp(viewer));
3632: /* read matrix header */
3633: PetscCall(PetscViewerBinaryRead(viewer, header, 4, NULL, PETSC_INT));
3634: PetscCheck(header[0] == MAT_FILE_CLASSID, PETSC_COMM_SELF, PETSC_ERR_FILE_UNEXPECTED, "Not a matrix object in file");
3635: M = header[1];
3636: N = header[2];
3637: nz = header[3];
3638: PetscCheck(M >= 0, PetscObjectComm((PetscObject)viewer), PETSC_ERR_FILE_UNEXPECTED, "Matrix row size (%" PetscInt_FMT ") in file is negative", M);
3639: PetscCheck(N >= 0, PetscObjectComm((PetscObject)viewer), PETSC_ERR_FILE_UNEXPECTED, "Matrix column size (%" PetscInt_FMT ") in file is negative", N);
3640: PetscCheck(nz >= 0, PETSC_COMM_SELF, PETSC_ERR_FILE_UNEXPECTED, "Matrix stored in special format on disk, cannot load as SeqBAIJ");
3642: /* set block sizes from the viewer's .info file */
3643: PetscCall(MatLoad_Binary_BlockSizes(mat, viewer));
3644: /* set local and global sizes if not set already */
3645: if (mat->rmap->n < 0) mat->rmap->n = M;
3646: if (mat->cmap->n < 0) mat->cmap->n = N;
3647: if (mat->rmap->N < 0) mat->rmap->N = M;
3648: if (mat->cmap->N < 0) mat->cmap->N = N;
3649: PetscCall(PetscLayoutSetUp(mat->rmap));
3650: PetscCall(PetscLayoutSetUp(mat->cmap));
3652: /* check if the matrix sizes are correct */
3653: PetscCall(MatGetSize(mat, &rows, &cols));
3654: PetscCheck(M == rows && N == cols, PETSC_COMM_SELF, PETSC_ERR_FILE_UNEXPECTED, "Matrix in file of different sizes (%" PetscInt_FMT ", %" PetscInt_FMT ") than the input matrix (%" PetscInt_FMT ", %" PetscInt_FMT ")", M, N, rows, cols);
3655: PetscCall(MatGetBlockSize(mat, &bs));
3656: PetscCall(MatGetLocalSize(mat, &m, &n));
3657: mbs = m / bs;
3658: nbs = n / bs;
3660: /* read in row lengths, column indices and nonzero values */
3661: PetscCall(PetscMalloc1(m + 1, &rowidxs));
3662: PetscCall(PetscViewerBinaryRead(viewer, rowidxs + 1, m, NULL, PETSC_INT));
3663: rowidxs[0] = 0;
3664: for (i = 0; i < m; i++) rowidxs[i + 1] += rowidxs[i];
3665: sum = rowidxs[m];
3666: PetscCheck(sum == nz, PETSC_COMM_SELF, PETSC_ERR_FILE_UNEXPECTED, "Inconsistent matrix data in file: nonzeros = %" PetscInt_FMT ", sum-row-lengths = %" PetscInt_FMT, nz, sum);
3668: /* read in column indices and nonzero values */
3669: PetscCall(PetscMalloc2(rowidxs[m], &colidxs, nz, &matvals));
3670: PetscCall(PetscViewerBinaryRead(viewer, colidxs, rowidxs[m], NULL, PETSC_INT));
3671: PetscCall(PetscViewerBinaryRead(viewer, matvals, rowidxs[m], NULL, PETSC_SCALAR));
3673: { /* preallocate matrix storage */
3674: PetscBT bt; /* helper bit set to count nonzeros */
3675: PetscInt *nnz;
3676: PetscBool sbaij;
3678: PetscCall(PetscBTCreate(nbs, &bt));
3679: PetscCall(PetscCalloc1(mbs, &nnz));
3680: PetscCall(PetscObjectTypeCompare((PetscObject)mat, MATSEQSBAIJ, &sbaij));
3681: for (i = 0; i < mbs; i++) {
3682: PetscCall(PetscBTMemzero(nbs, bt));
3683: for (k = 0; k < bs; k++) {
3684: PetscInt row = bs * i + k;
3685: for (j = rowidxs[row]; j < rowidxs[row + 1]; j++) {
3686: PetscInt col = colidxs[j];
3687: if (!sbaij || col >= row)
3688: if (!PetscBTLookupSet(bt, col / bs)) nnz[i]++;
3689: }
3690: }
3691: }
3692: PetscCall(PetscBTDestroy(&bt));
3693: PetscCall(MatSeqBAIJSetPreallocation(mat, bs, 0, nnz));
3694: PetscCall(MatSeqSBAIJSetPreallocation(mat, bs, 0, nnz));
3695: PetscCall(PetscFree(nnz));
3696: }
3698: /* store matrix values */
3699: for (i = 0; i < m; i++) {
3700: PetscInt row = i, s = rowidxs[i], e = rowidxs[i + 1];
3701: PetscCall((*mat->ops->setvalues)(mat, 1, &row, e - s, colidxs + s, matvals + s, INSERT_VALUES));
3702: }
3704: PetscCall(PetscFree(rowidxs));
3705: PetscCall(PetscFree2(colidxs, matvals));
3706: PetscCall(MatAssemblyBegin(mat, MAT_FINAL_ASSEMBLY));
3707: PetscCall(MatAssemblyEnd(mat, MAT_FINAL_ASSEMBLY));
3708: PetscFunctionReturn(PETSC_SUCCESS);
3709: }
3711: PetscErrorCode MatLoad_SeqBAIJ(Mat mat, PetscViewer viewer)
3712: {
3713: PetscBool isbinary;
3715: PetscFunctionBegin;
3716: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERBINARY, &isbinary));
3717: PetscCheck(isbinary, PetscObjectComm((PetscObject)viewer), PETSC_ERR_SUP, "Viewer type %s not yet supported for reading %s matrices", ((PetscObject)viewer)->type_name, ((PetscObject)mat)->type_name);
3718: PetscCall(MatLoad_SeqBAIJ_Binary(mat, viewer));
3719: PetscFunctionReturn(PETSC_SUCCESS);
3720: }
3722: /*@C
3723: MatCreateSeqBAIJ - Creates a sparse matrix in `MATSEQAIJ` (block
3724: compressed row) format. For good matrix assembly performance the
3725: user should preallocate the matrix storage by setting the parameter `nz`
3726: (or the array `nnz`).
3728: Collective
3730: Input Parameters:
3731: + comm - MPI communicator, set to `PETSC_COMM_SELF`
3732: . bs - size of block, the blocks are ALWAYS square. One can use `MatSetBlockSizes()` to set a different row and column blocksize but the row
3733: blocksize always defines the size of the blocks. The column blocksize sets the blocksize of the vectors obtained with `MatCreateVecs()`
3734: . m - number of rows
3735: . n - number of columns
3736: . nz - number of nonzero blocks per block row (same for all rows)
3737: - nnz - array containing the number of nonzero blocks in the various block rows
3738: (possibly different for each block row) or `NULL`
3740: Output Parameter:
3741: . A - the matrix
3743: Options Database Keys:
3744: + -mat_no_unroll - uses code that does not unroll the loops in the block calculations (much slower)
3745: - -mat_block_size - size of the blocks to use
3747: Level: intermediate
3749: Notes:
3750: It is recommended that one use the `MatCreate()`, `MatSetType()` and/or `MatSetFromOptions()`,
3751: MatXXXXSetPreallocation() paradigm instead of this routine directly.
3752: [MatXXXXSetPreallocation() is, for example, `MatSeqAIJSetPreallocation()`]
3754: The number of rows and columns must be divisible by blocksize.
3756: If the `nnz` parameter is given then the `nz` parameter is ignored
3758: A nonzero block is any block that as 1 or more nonzeros in it
3760: The `MATSEQBAIJ` format is fully compatible with standard Fortran
3761: storage. That is, the stored row and column indices can begin at
3762: either one (as in Fortran) or zero.
3764: Specify the preallocated storage with either `nz` or `nnz` (not both).
3765: Set `nz` = `PETSC_DEFAULT` and `nnz` = `NULL` for PETSc to control dynamic memory
3766: allocation. See [Sparse Matrices](sec_matsparse) for details.
3767: matrices.
3769: .seealso: [](ch_matrices), `Mat`, [Sparse Matrices](sec_matsparse), `MatCreate()`, `MatCreateSeqAIJ()`, `MatSetValues()`, `MatCreateBAIJ()`
3770: @*/
3771: PetscErrorCode MatCreateSeqBAIJ(MPI_Comm comm, PetscInt bs, PetscInt m, PetscInt n, PetscInt nz, const PetscInt nnz[], Mat *A)
3772: {
3773: PetscFunctionBegin;
3774: PetscCall(MatCreate(comm, A));
3775: PetscCall(MatSetSizes(*A, m, n, m, n));
3776: PetscCall(MatSetType(*A, MATSEQBAIJ));
3777: PetscCall(MatSeqBAIJSetPreallocation(*A, bs, nz, (PetscInt *)nnz));
3778: PetscFunctionReturn(PETSC_SUCCESS);
3779: }
3781: /*@C
3782: MatSeqBAIJSetPreallocation - Sets the block size and expected nonzeros
3783: per row in the matrix. For good matrix assembly performance the
3784: user should preallocate the matrix storage by setting the parameter `nz`
3785: (or the array `nnz`).
3787: Collective
3789: Input Parameters:
3790: + B - the matrix
3791: . bs - size of block, the blocks are ALWAYS square. One can use `MatSetBlockSizes()` to set a different row and column blocksize but the row
3792: blocksize always defines the size of the blocks. The column blocksize sets the blocksize of the vectors obtained with `MatCreateVecs()`
3793: . nz - number of block nonzeros per block row (same for all rows)
3794: - nnz - array containing the number of block nonzeros in the various block rows
3795: (possibly different for each block row) or `NULL`
3797: Options Database Keys:
3798: + -mat_no_unroll - uses code that does not unroll the loops in the block calculations (much slower)
3799: - -mat_block_size - size of the blocks to use
3801: Level: intermediate
3803: Notes:
3804: If the `nnz` parameter is given then the `nz` parameter is ignored
3806: You can call `MatGetInfo()` to get information on how effective the preallocation was;
3807: for example the fields mallocs,nz_allocated,nz_used,nz_unneeded;
3808: You can also run with the option `-info` and look for messages with the string
3809: malloc in them to see if additional memory allocation was needed.
3811: The `MATSEQBAIJ` format is fully compatible with standard Fortran
3812: storage. That is, the stored row and column indices can begin at
3813: either one (as in Fortran) or zero.
3815: Specify the preallocated storage with either nz or nnz (not both).
3816: Set `nz` = `PETSC_DEFAULT` and `nnz` = `NULL` for PETSc to control dynamic memory
3817: allocation. See [Sparse Matrices](sec_matsparse) for details.
3819: .seealso: [](ch_matrices), `Mat`, [Sparse Matrices](sec_matsparse), `MatCreate()`, `MatCreateSeqAIJ()`, `MatSetValues()`, `MatCreateBAIJ()`, `MatGetInfo()`
3820: @*/
3821: PetscErrorCode MatSeqBAIJSetPreallocation(Mat B, PetscInt bs, PetscInt nz, const PetscInt nnz[])
3822: {
3823: PetscFunctionBegin;
3827: PetscTryMethod(B, "MatSeqBAIJSetPreallocation_C", (Mat, PetscInt, PetscInt, const PetscInt[]), (B, bs, nz, nnz));
3828: PetscFunctionReturn(PETSC_SUCCESS);
3829: }
3831: /*@C
3832: MatSeqBAIJSetPreallocationCSR - Creates a sparse sequential matrix in `MATSEQBAIJ` format using the given nonzero structure and (optional) numerical values
3834: Collective
3836: Input Parameters:
3837: + B - the matrix
3838: . bs - the blocksize
3839: . i - the indices into `j` for the start of each local row (starts with zero)
3840: . j - the column indices for each local row (starts with zero) these must be sorted for each row
3841: - v - optional values in the matrix
3843: Level: advanced
3845: Notes:
3846: The order of the entries in values is specified by the `MatOption` `MAT_ROW_ORIENTED`. For example, C programs
3847: may want to use the default `MAT_ROW_ORIENTED` of `PETSC_TRUE` and use an array v[nnz][bs][bs] where the second index is
3848: over rows within a block and the last index is over columns within a block row. Fortran programs will likely set
3849: `MAT_ROW_ORIENTED` of `PETSC_FALSE` and use a Fortran array v(bs,bs,nnz) in which the first index is over rows within a
3850: block column and the second index is over columns within a block.
3852: Though this routine has Preallocation() in the name it also sets the exact nonzero locations of the matrix entries and usually the numerical values as well
3854: .seealso: [](ch_matrices), `Mat`, `MatCreate()`, `MatCreateSeqBAIJ()`, `MatSetValues()`, `MatSeqBAIJSetPreallocation()`, `MATSEQBAIJ`
3855: @*/
3856: PetscErrorCode MatSeqBAIJSetPreallocationCSR(Mat B, PetscInt bs, const PetscInt i[], const PetscInt j[], const PetscScalar v[])
3857: {
3858: PetscFunctionBegin;
3862: PetscTryMethod(B, "MatSeqBAIJSetPreallocationCSR_C", (Mat, PetscInt, const PetscInt[], const PetscInt[], const PetscScalar[]), (B, bs, i, j, v));
3863: PetscFunctionReturn(PETSC_SUCCESS);
3864: }
3866: /*@
3867: MatCreateSeqBAIJWithArrays - Creates a `MATSEQBAIJ` matrix using matrix elements provided by the user.
3869: Collective
3871: Input Parameters:
3872: + comm - must be an MPI communicator of size 1
3873: . bs - size of block
3874: . m - number of rows
3875: . n - number of columns
3876: . i - row indices; that is i[0] = 0, i[row] = i[row-1] + number of elements in that row block row of the matrix
3877: . j - column indices
3878: - a - matrix values
3880: Output Parameter:
3881: . mat - the matrix
3883: Level: advanced
3885: Notes:
3886: The `i`, `j`, and `a` arrays are not copied by this routine, the user must free these arrays
3887: once the matrix is destroyed
3889: You cannot set new nonzero locations into this matrix, that will generate an error.
3891: The `i` and `j` indices are 0 based
3893: When block size is greater than 1 the matrix values must be stored using the `MATSEQBAIJ` storage format
3895: The order of the entries in values is the same as the block compressed sparse row storage format; that is, it is
3896: the same as a three dimensional array in Fortran values(bs,bs,nnz) that contains the first column of the first
3897: block, followed by the second column of the first block etc etc. That is, the blocks are contiguous in memory
3898: with column-major ordering within blocks.
3900: .seealso: [](ch_matrices), `Mat`, `MatCreate()`, `MatCreateBAIJ()`, `MatCreateSeqBAIJ()`
3901: @*/
3902: PetscErrorCode MatCreateSeqBAIJWithArrays(MPI_Comm comm, PetscInt bs, PetscInt m, PetscInt n, PetscInt i[], PetscInt j[], PetscScalar a[], Mat *mat)
3903: {
3904: Mat_SeqBAIJ *baij;
3906: PetscFunctionBegin;
3907: PetscCheck(bs == 1, PETSC_COMM_SELF, PETSC_ERR_SUP, "block size %" PetscInt_FMT " > 1 is not supported yet", bs);
3908: if (m > 0) PetscCheck(i[0] == 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "i (row indices) must start with 0");
3910: PetscCall(MatCreate(comm, mat));
3911: PetscCall(MatSetSizes(*mat, m, n, m, n));
3912: PetscCall(MatSetType(*mat, MATSEQBAIJ));
3913: PetscCall(MatSeqBAIJSetPreallocation(*mat, bs, MAT_SKIP_ALLOCATION, NULL));
3914: baij = (Mat_SeqBAIJ *)(*mat)->data;
3915: PetscCall(PetscMalloc2(m, &baij->imax, m, &baij->ilen));
3917: baij->i = i;
3918: baij->j = j;
3919: baij->a = a;
3921: baij->singlemalloc = PETSC_FALSE;
3922: baij->nonew = -1; /*this indicates that inserting a new value in the matrix that generates a new nonzero is an error*/
3923: baij->free_a = PETSC_FALSE;
3924: baij->free_ij = PETSC_FALSE;
3925: baij->free_imax_ilen = PETSC_TRUE;
3927: for (PetscInt ii = 0; ii < m; ii++) {
3928: const PetscInt row_len = i[ii + 1] - i[ii];
3930: baij->ilen[ii] = baij->imax[ii] = row_len;
3931: PetscCheck(row_len >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Negative row length in i (row indices) row = %" PetscInt_FMT " length = %" PetscInt_FMT, ii, row_len);
3932: }
3933: if (PetscDefined(USE_DEBUG)) {
3934: for (PetscInt ii = 0; ii < baij->i[m]; ii++) {
3935: PetscCheck(j[ii] >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Negative column index at location = %" PetscInt_FMT " index = %" PetscInt_FMT, ii, j[ii]);
3936: PetscCheck(j[ii] <= n - 1, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Column index to large at location = %" PetscInt_FMT " index = %" PetscInt_FMT, ii, j[ii]);
3937: }
3938: }
3940: PetscCall(MatAssemblyBegin(*mat, MAT_FINAL_ASSEMBLY));
3941: PetscCall(MatAssemblyEnd(*mat, MAT_FINAL_ASSEMBLY));
3942: PetscFunctionReturn(PETSC_SUCCESS);
3943: }
3945: PetscErrorCode MatCreateMPIMatConcatenateSeqMat_SeqBAIJ(MPI_Comm comm, Mat inmat, PetscInt n, MatReuse scall, Mat *outmat)
3946: {
3947: PetscFunctionBegin;
3948: PetscCall(MatCreateMPIMatConcatenateSeqMat_MPIBAIJ(comm, inmat, n, scall, outmat));
3949: PetscFunctionReturn(PETSC_SUCCESS);
3950: }