xref: /petsc/src/ksp/pc/impls/pbjacobi/pbjacobi.c (revision b8d709ab44c354dba7591c0191c0984ef3bb0cd5)
1 
2 /*
3    Include files needed for the PBJacobi preconditioner:
4      pcimpl.h - private include file intended for use by all preconditioners
5 */
6 
7 #include <petsc/private/pcimpl.h> /*I "petscpc.h" I*/
8 
9 /*
10    Private context (data structure) for the PBJacobi preconditioner.
11 */
12 typedef struct {
13   const MatScalar *diag;
14   PetscInt         bs, mbs;
15 } PC_PBJacobi;
16 
17 static PetscErrorCode PCApply_PBJacobi_1(PC pc, Vec x, Vec y)
18 {
19   PC_PBJacobi       *jac = (PC_PBJacobi *)pc->data;
20   PetscInt           i, m = jac->mbs;
21   const MatScalar   *diag = jac->diag;
22   const PetscScalar *xx;
23   PetscScalar       *yy;
24 
25   PetscFunctionBegin;
26   PetscCall(VecGetArrayRead(x, &xx));
27   PetscCall(VecGetArray(y, &yy));
28   for (i = 0; i < m; i++) yy[i] = diag[i] * xx[i];
29   PetscCall(VecRestoreArrayRead(x, &xx));
30   PetscCall(VecRestoreArray(y, &yy));
31   PetscCall(PetscLogFlops(m));
32   PetscFunctionReturn(0);
33 }
34 
35 static PetscErrorCode PCApply_PBJacobi_2(PC pc, Vec x, Vec y)
36 {
37   PC_PBJacobi       *jac = (PC_PBJacobi *)pc->data;
38   PetscInt           i, m = jac->mbs;
39   const MatScalar   *diag = jac->diag;
40   PetscScalar        x0, x1, *yy;
41   const PetscScalar *xx;
42 
43   PetscFunctionBegin;
44   PetscCall(VecGetArrayRead(x, &xx));
45   PetscCall(VecGetArray(y, &yy));
46   for (i = 0; i < m; i++) {
47     x0            = xx[2 * i];
48     x1            = xx[2 * i + 1];
49     yy[2 * i]     = diag[0] * x0 + diag[2] * x1;
50     yy[2 * i + 1] = diag[1] * x0 + diag[3] * x1;
51     diag += 4;
52   }
53   PetscCall(VecRestoreArrayRead(x, &xx));
54   PetscCall(VecRestoreArray(y, &yy));
55   PetscCall(PetscLogFlops(6.0 * m));
56   PetscFunctionReturn(0);
57 }
58 static PetscErrorCode PCApply_PBJacobi_3(PC pc, Vec x, Vec y)
59 {
60   PC_PBJacobi       *jac = (PC_PBJacobi *)pc->data;
61   PetscInt           i, m = jac->mbs;
62   const MatScalar   *diag = jac->diag;
63   PetscScalar        x0, x1, x2, *yy;
64   const PetscScalar *xx;
65 
66   PetscFunctionBegin;
67   PetscCall(VecGetArrayRead(x, &xx));
68   PetscCall(VecGetArray(y, &yy));
69   for (i = 0; i < m; i++) {
70     x0 = xx[3 * i];
71     x1 = xx[3 * i + 1];
72     x2 = xx[3 * i + 2];
73 
74     yy[3 * i]     = diag[0] * x0 + diag[3] * x1 + diag[6] * x2;
75     yy[3 * i + 1] = diag[1] * x0 + diag[4] * x1 + diag[7] * x2;
76     yy[3 * i + 2] = diag[2] * x0 + diag[5] * x1 + diag[8] * x2;
77     diag += 9;
78   }
79   PetscCall(VecRestoreArrayRead(x, &xx));
80   PetscCall(VecRestoreArray(y, &yy));
81   PetscCall(PetscLogFlops(15.0 * m));
82   PetscFunctionReturn(0);
83 }
84 static PetscErrorCode PCApply_PBJacobi_4(PC pc, Vec x, Vec y)
85 {
86   PC_PBJacobi       *jac = (PC_PBJacobi *)pc->data;
87   PetscInt           i, m = jac->mbs;
88   const MatScalar   *diag = jac->diag;
89   PetscScalar        x0, x1, x2, x3, *yy;
90   const PetscScalar *xx;
91 
92   PetscFunctionBegin;
93   PetscCall(VecGetArrayRead(x, &xx));
94   PetscCall(VecGetArray(y, &yy));
95   for (i = 0; i < m; i++) {
96     x0 = xx[4 * i];
97     x1 = xx[4 * i + 1];
98     x2 = xx[4 * i + 2];
99     x3 = xx[4 * i + 3];
100 
101     yy[4 * i]     = diag[0] * x0 + diag[4] * x1 + diag[8] * x2 + diag[12] * x3;
102     yy[4 * i + 1] = diag[1] * x0 + diag[5] * x1 + diag[9] * x2 + diag[13] * x3;
103     yy[4 * i + 2] = diag[2] * x0 + diag[6] * x1 + diag[10] * x2 + diag[14] * x3;
104     yy[4 * i + 3] = diag[3] * x0 + diag[7] * x1 + diag[11] * x2 + diag[15] * x3;
105     diag += 16;
106   }
107   PetscCall(VecRestoreArrayRead(x, &xx));
108   PetscCall(VecRestoreArray(y, &yy));
109   PetscCall(PetscLogFlops(28.0 * m));
110   PetscFunctionReturn(0);
111 }
112 static PetscErrorCode PCApply_PBJacobi_5(PC pc, Vec x, Vec y)
113 {
114   PC_PBJacobi       *jac = (PC_PBJacobi *)pc->data;
115   PetscInt           i, m = jac->mbs;
116   const MatScalar   *diag = jac->diag;
117   PetscScalar        x0, x1, x2, x3, x4, *yy;
118   const PetscScalar *xx;
119 
120   PetscFunctionBegin;
121   PetscCall(VecGetArrayRead(x, &xx));
122   PetscCall(VecGetArray(y, &yy));
123   for (i = 0; i < m; i++) {
124     x0 = xx[5 * i];
125     x1 = xx[5 * i + 1];
126     x2 = xx[5 * i + 2];
127     x3 = xx[5 * i + 3];
128     x4 = xx[5 * i + 4];
129 
130     yy[5 * i]     = diag[0] * x0 + diag[5] * x1 + diag[10] * x2 + diag[15] * x3 + diag[20] * x4;
131     yy[5 * i + 1] = diag[1] * x0 + diag[6] * x1 + diag[11] * x2 + diag[16] * x3 + diag[21] * x4;
132     yy[5 * i + 2] = diag[2] * x0 + diag[7] * x1 + diag[12] * x2 + diag[17] * x3 + diag[22] * x4;
133     yy[5 * i + 3] = diag[3] * x0 + diag[8] * x1 + diag[13] * x2 + diag[18] * x3 + diag[23] * x4;
134     yy[5 * i + 4] = diag[4] * x0 + diag[9] * x1 + diag[14] * x2 + diag[19] * x3 + diag[24] * x4;
135     diag += 25;
136   }
137   PetscCall(VecRestoreArrayRead(x, &xx));
138   PetscCall(VecRestoreArray(y, &yy));
139   PetscCall(PetscLogFlops(45.0 * m));
140   PetscFunctionReturn(0);
141 }
142 static PetscErrorCode PCApply_PBJacobi_6(PC pc, Vec x, Vec y)
143 {
144   PC_PBJacobi       *jac = (PC_PBJacobi *)pc->data;
145   PetscInt           i, m = jac->mbs;
146   const MatScalar   *diag = jac->diag;
147   PetscScalar        x0, x1, x2, x3, x4, x5, *yy;
148   const PetscScalar *xx;
149 
150   PetscFunctionBegin;
151   PetscCall(VecGetArrayRead(x, &xx));
152   PetscCall(VecGetArray(y, &yy));
153   for (i = 0; i < m; i++) {
154     x0 = xx[6 * i];
155     x1 = xx[6 * i + 1];
156     x2 = xx[6 * i + 2];
157     x3 = xx[6 * i + 3];
158     x4 = xx[6 * i + 4];
159     x5 = xx[6 * i + 5];
160 
161     yy[6 * i]     = diag[0] * x0 + diag[6] * x1 + diag[12] * x2 + diag[18] * x3 + diag[24] * x4 + diag[30] * x5;
162     yy[6 * i + 1] = diag[1] * x0 + diag[7] * x1 + diag[13] * x2 + diag[19] * x3 + diag[25] * x4 + diag[31] * x5;
163     yy[6 * i + 2] = diag[2] * x0 + diag[8] * x1 + diag[14] * x2 + diag[20] * x3 + diag[26] * x4 + diag[32] * x5;
164     yy[6 * i + 3] = diag[3] * x0 + diag[9] * x1 + diag[15] * x2 + diag[21] * x3 + diag[27] * x4 + diag[33] * x5;
165     yy[6 * i + 4] = diag[4] * x0 + diag[10] * x1 + diag[16] * x2 + diag[22] * x3 + diag[28] * x4 + diag[34] * x5;
166     yy[6 * i + 5] = diag[5] * x0 + diag[11] * x1 + diag[17] * x2 + diag[23] * x3 + diag[29] * x4 + diag[35] * x5;
167     diag += 36;
168   }
169   PetscCall(VecRestoreArrayRead(x, &xx));
170   PetscCall(VecRestoreArray(y, &yy));
171   PetscCall(PetscLogFlops(66.0 * m));
172   PetscFunctionReturn(0);
173 }
174 static PetscErrorCode PCApply_PBJacobi_7(PC pc, Vec x, Vec y)
175 {
176   PC_PBJacobi       *jac = (PC_PBJacobi *)pc->data;
177   PetscInt           i, m = jac->mbs;
178   const MatScalar   *diag = jac->diag;
179   PetscScalar        x0, x1, x2, x3, x4, x5, x6, *yy;
180   const PetscScalar *xx;
181 
182   PetscFunctionBegin;
183   PetscCall(VecGetArrayRead(x, &xx));
184   PetscCall(VecGetArray(y, &yy));
185   for (i = 0; i < m; i++) {
186     x0 = xx[7 * i];
187     x1 = xx[7 * i + 1];
188     x2 = xx[7 * i + 2];
189     x3 = xx[7 * i + 3];
190     x4 = xx[7 * i + 4];
191     x5 = xx[7 * i + 5];
192     x6 = xx[7 * i + 6];
193 
194     yy[7 * i]     = diag[0] * x0 + diag[7] * x1 + diag[14] * x2 + diag[21] * x3 + diag[28] * x4 + diag[35] * x5 + diag[42] * x6;
195     yy[7 * i + 1] = diag[1] * x0 + diag[8] * x1 + diag[15] * x2 + diag[22] * x3 + diag[29] * x4 + diag[36] * x5 + diag[43] * x6;
196     yy[7 * i + 2] = diag[2] * x0 + diag[9] * x1 + diag[16] * x2 + diag[23] * x3 + diag[30] * x4 + diag[37] * x5 + diag[44] * x6;
197     yy[7 * i + 3] = diag[3] * x0 + diag[10] * x1 + diag[17] * x2 + diag[24] * x3 + diag[31] * x4 + diag[38] * x5 + diag[45] * x6;
198     yy[7 * i + 4] = diag[4] * x0 + diag[11] * x1 + diag[18] * x2 + diag[25] * x3 + diag[32] * x4 + diag[39] * x5 + diag[46] * x6;
199     yy[7 * i + 5] = diag[5] * x0 + diag[12] * x1 + diag[19] * x2 + diag[26] * x3 + diag[33] * x4 + diag[40] * x5 + diag[47] * x6;
200     yy[7 * i + 6] = diag[6] * x0 + diag[13] * x1 + diag[20] * x2 + diag[27] * x3 + diag[34] * x4 + diag[41] * x5 + diag[48] * x6;
201     diag += 49;
202   }
203   PetscCall(VecRestoreArrayRead(x, &xx));
204   PetscCall(VecRestoreArray(y, &yy));
205   PetscCall(PetscLogFlops(91.0 * m)); /* 2*bs2 - bs */
206   PetscFunctionReturn(0);
207 }
208 static PetscErrorCode PCApply_PBJacobi_N(PC pc, Vec x, Vec y)
209 {
210   PC_PBJacobi       *jac = (PC_PBJacobi *)pc->data;
211   PetscInt           i, ib, jb;
212   const PetscInt     m    = jac->mbs;
213   const PetscInt     bs   = jac->bs;
214   const MatScalar   *diag = jac->diag;
215   PetscScalar       *yy;
216   const PetscScalar *xx;
217 
218   PetscFunctionBegin;
219   PetscCall(VecGetArrayRead(x, &xx));
220   PetscCall(VecGetArray(y, &yy));
221   for (i = 0; i < m; i++) {
222     for (ib = 0; ib < bs; ib++) {
223       PetscScalar rowsum = 0;
224       for (jb = 0; jb < bs; jb++) rowsum += diag[ib + jb * bs] * xx[bs * i + jb];
225       yy[bs * i + ib] = rowsum;
226     }
227     diag += bs * bs;
228   }
229   PetscCall(VecRestoreArrayRead(x, &xx));
230   PetscCall(VecRestoreArray(y, &yy));
231   PetscCall(PetscLogFlops((2.0 * bs * bs - bs) * m)); /* 2*bs2 - bs */
232   PetscFunctionReturn(0);
233 }
234 
235 static PetscErrorCode PCApplyTranspose_PBJacobi_N(PC pc, Vec x, Vec y)
236 {
237   PC_PBJacobi       *jac = (PC_PBJacobi *)pc->data;
238   PetscInt           i, j, k, m = jac->mbs, bs = jac->bs;
239   const MatScalar   *diag = jac->diag;
240   const PetscScalar *xx;
241   PetscScalar       *yy;
242 
243   PetscFunctionBegin;
244   PetscCall(VecGetArrayRead(x, &xx));
245   PetscCall(VecGetArray(y, &yy));
246   for (i = 0; i < m; i++) {
247     for (j = 0; j < bs; j++) yy[i * bs + j] = 0.;
248     for (j = 0; j < bs; j++) {
249       for (k = 0; k < bs; k++) yy[i * bs + k] += diag[k * bs + j] * xx[i * bs + j];
250     }
251     diag += bs * bs;
252   }
253   PetscCall(VecRestoreArrayRead(x, &xx));
254   PetscCall(VecRestoreArray(y, &yy));
255   PetscCall(PetscLogFlops(m * bs * (2 * bs - 1)));
256   PetscFunctionReturn(0);
257 }
258 
259 static PetscErrorCode PCSetUp_PBJacobi(PC pc)
260 {
261   PC_PBJacobi   *jac = (PC_PBJacobi *)pc->data;
262   Mat            A   = pc->pmat;
263   MatFactorError err;
264   PetscInt       nlocal;
265 
266   PetscFunctionBegin;
267   PetscCall(MatInvertBlockDiagonal(A, &jac->diag));
268   PetscCall(MatFactorGetError(A, &err));
269   if (err) pc->failedreason = (PCFailedReason)err;
270 
271   PetscCall(MatGetBlockSize(A, &jac->bs));
272   PetscCall(MatGetLocalSize(A, &nlocal, NULL));
273   jac->mbs = nlocal / jac->bs;
274   switch (jac->bs) {
275   case 1:
276     pc->ops->apply = PCApply_PBJacobi_1;
277     break;
278   case 2:
279     pc->ops->apply = PCApply_PBJacobi_2;
280     break;
281   case 3:
282     pc->ops->apply = PCApply_PBJacobi_3;
283     break;
284   case 4:
285     pc->ops->apply = PCApply_PBJacobi_4;
286     break;
287   case 5:
288     pc->ops->apply = PCApply_PBJacobi_5;
289     break;
290   case 6:
291     pc->ops->apply = PCApply_PBJacobi_6;
292     break;
293   case 7:
294     pc->ops->apply = PCApply_PBJacobi_7;
295     break;
296   default:
297     pc->ops->apply = PCApply_PBJacobi_N;
298     break;
299   }
300   pc->ops->applytranspose = PCApplyTranspose_PBJacobi_N;
301   PetscFunctionReturn(0);
302 }
303 
304 static PetscErrorCode PCDestroy_PBJacobi(PC pc)
305 {
306   PetscFunctionBegin;
307   /*
308       Free the private data structure that was hanging off the PC
309   */
310   PetscCall(PetscFree(pc->data));
311   PetscFunctionReturn(0);
312 }
313 
314 static PetscErrorCode PCView_PBJacobi(PC pc, PetscViewer viewer)
315 {
316   PC_PBJacobi *jac = (PC_PBJacobi *)pc->data;
317   PetscBool    iascii;
318 
319   PetscFunctionBegin;
320   PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &iascii));
321   if (iascii) PetscCall(PetscViewerASCIIPrintf(viewer, "  point-block size %" PetscInt_FMT "\n", jac->bs));
322   PetscFunctionReturn(0);
323 }
324 
325 /*MC
326      PCPBJACOBI - Point block Jacobi preconditioner
327 
328    Notes:
329     See `PCJACOBI` for diagonal Jacobi, `PCVPBJACOBI` for variable-size point block, and `PCBJACOBI` for large size blocks
330 
331    This works for `MATAIJ` and `MATBAIJ` matrices and uses the blocksize provided to the matrix
332 
333    Uses dense LU factorization with partial pivoting to invert the blocks; if a zero pivot
334    is detected a PETSc error is generated.
335 
336    Developer Notes:
337      This should support the `PCSetErrorIfFailure()` flag set to `PETSC_TRUE` to allow
338      the factorization to continue even after a zero pivot is found resulting in a Nan and hence
339      terminating `KSP` with a `KSP_DIVERGED_NANORINF` allowing
340      a nonlinear solver/ODE integrator to recover without stopping the program as currently happens.
341 
342      Perhaps should provide an option that allows generation of a valid preconditioner
343      even if a block is singular as the `PCJACOBI` does.
344 
345    Level: beginner
346 
347 .seealso: `PCCreate()`, `PCSetType()`, `PCType`, `PC`, `PCJACOBI`, `PCVPBJACOBI`, `PCBJACOBI`
348 M*/
349 
350 PETSC_EXTERN PetscErrorCode PCCreate_PBJacobi(PC pc)
351 {
352   PC_PBJacobi *jac;
353 
354   PetscFunctionBegin;
355   /*
356      Creates the private data structure for this preconditioner and
357      attach it to the PC object.
358   */
359   PetscCall(PetscNew(&jac));
360   pc->data = (void *)jac;
361 
362   /*
363      Initialize the pointers to vectors to ZERO; these will be used to store
364      diagonal entries of the matrix for fast preconditioner application.
365   */
366   jac->diag = NULL;
367 
368   /*
369       Set the pointers for the functions that are provided above.
370       Now when the user-level routines (such as PCApply(), PCDestroy(), etc.)
371       are called, they will automatically call these functions.  Note we
372       choose not to provide a couple of these functions since they are
373       not needed.
374   */
375   pc->ops->apply               = NULL; /*set depending on the block size */
376   pc->ops->applytranspose      = NULL;
377   pc->ops->setup               = PCSetUp_PBJacobi;
378   pc->ops->destroy             = PCDestroy_PBJacobi;
379   pc->ops->setfromoptions      = NULL;
380   pc->ops->view                = PCView_PBJacobi;
381   pc->ops->applyrichardson     = NULL;
382   pc->ops->applysymmetricleft  = NULL;
383   pc->ops->applysymmetricright = NULL;
384   PetscFunctionReturn(0);
385 }
386