Actual source code: mpisell.c
1: #include <../src/mat/impls/aij/mpi/mpiaij.h>
2: #include <../src/mat/impls/sell/mpi/mpisell.h>
3: #include <petsc/private/vecimpl.h>
4: #include <petsc/private/isimpl.h>
5: #include <petscblaslapack.h>
6: #include <petscsf.h>
8: /*MC
9: MATSELL - MATSELL = "sell" - A matrix type to be used for sparse matrices.
11: This matrix type is identical to `MATSEQSELL` when constructed with a single process communicator,
12: and `MATMPISELL` otherwise. As a result, for single process communicators,
13: `MatSeqSELLSetPreallocation()` is supported, and similarly `MatMPISELLSetPreallocation()` is supported
14: for communicators controlling multiple processes. It is recommended that you call both of
15: the above preallocation routines for simplicity.
17: Options Database Keys:
18: . -mat_type sell - sets the matrix type to `MATSELL` during a call to `MatSetFromOptions()`
20: Level: beginner
22: .seealso: `Mat`, `MATAIJ`, `MATBAIJ`, `MATSBAIJ`, `MatCreateSELL()`, `MatCreateSeqSELL()`, `MATSEQSELL`, `MATMPISELL`
23: M*/
25: static PetscErrorCode MatDiagonalSet_MPISELL(Mat Y, Vec D, InsertMode is)
26: {
27: Mat_MPISELL *sell = (Mat_MPISELL *)Y->data;
29: PetscFunctionBegin;
30: if (Y->assembled && Y->rmap->rstart == Y->cmap->rstart && Y->rmap->rend == Y->cmap->rend) {
31: PetscCall(MatDiagonalSet(sell->A, D, is));
32: } else {
33: PetscCall(MatDiagonalSet_Default(Y, D, is));
34: }
35: PetscFunctionReturn(PETSC_SUCCESS);
36: }
38: /*
39: Local utility routine that creates a mapping from the global column
40: number to the local number in the off-diagonal part of the local
41: storage of the matrix. When PETSC_USE_CTABLE is used this is scalable at
42: a slightly higher hash table cost; without it it is not scalable (each processor
43: has an order N integer array but is fast to access.
44: */
45: PetscErrorCode MatCreateColmap_MPISELL_Private(Mat mat)
46: {
47: Mat_MPISELL *sell = (Mat_MPISELL *)mat->data;
48: PetscInt n = sell->B->cmap->n, i;
50: PetscFunctionBegin;
51: PetscCheck(sell->garray, PETSC_COMM_SELF, PETSC_ERR_PLIB, "MPISELL Matrix was assembled but is missing garray");
52: #if defined(PETSC_USE_CTABLE)
53: PetscCall(PetscHMapICreateWithSize(n, &sell->colmap));
54: for (i = 0; i < n; i++) PetscCall(PetscHMapISet(sell->colmap, sell->garray[i] + 1, i + 1));
55: #else
56: PetscCall(PetscCalloc1(mat->cmap->N + 1, &sell->colmap));
57: for (i = 0; i < n; i++) sell->colmap[sell->garray[i]] = i + 1;
58: #endif
59: PetscFunctionReturn(PETSC_SUCCESS);
60: }
62: #define MatSetValues_SeqSELL_A_Private(row, col, value, addv, orow, ocol) \
63: { \
64: if (col <= lastcol1) low1 = 0; \
65: else high1 = nrow1; \
66: lastcol1 = col; \
67: while (high1 - low1 > 5) { \
68: t = (low1 + high1) / 2; \
69: if (cp1[sliceheight * t] > col) high1 = t; \
70: else low1 = t; \
71: } \
72: for (_i = low1; _i < high1; _i++) { \
73: if (cp1[sliceheight * _i] > col) break; \
74: if (cp1[sliceheight * _i] == col) { \
75: if (addv == ADD_VALUES) vp1[sliceheight * _i] += value; \
76: else vp1[sliceheight * _i] = value; \
77: inserted = PETSC_TRUE; \
78: goto a_noinsert; \
79: } \
80: } \
81: if (value == 0.0 && ignorezeroentries) { \
82: low1 = 0; \
83: high1 = nrow1; \
84: goto a_noinsert; \
85: } \
86: if (nonew == 1) { \
87: low1 = 0; \
88: high1 = nrow1; \
89: goto a_noinsert; \
90: } \
91: PetscCheck(nonew != -1, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Inserting a new nonzero at global row/column (%" PetscInt_FMT ", %" PetscInt_FMT ") into matrix", orow, ocol); \
92: MatSeqXSELLReallocateSELL(A, am, 1, nrow1, a->sliidx, a->sliceheight, row / sliceheight, row, col, a->colidx, a->val, cp1, vp1, nonew, MatScalar); \
93: /* shift up all the later entries in this row */ \
94: for (ii = nrow1 - 1; ii >= _i; ii--) { \
95: cp1[sliceheight * (ii + 1)] = cp1[sliceheight * ii]; \
96: vp1[sliceheight * (ii + 1)] = vp1[sliceheight * ii]; \
97: } \
98: cp1[sliceheight * _i] = col; \
99: vp1[sliceheight * _i] = value; \
100: a->nz++; \
101: nrow1++; \
102: a_noinsert:; \
103: a->rlen[row] = nrow1; \
104: }
106: #define MatSetValues_SeqSELL_B_Private(row, col, value, addv, orow, ocol) \
107: { \
108: if (col <= lastcol2) low2 = 0; \
109: else high2 = nrow2; \
110: lastcol2 = col; \
111: while (high2 - low2 > 5) { \
112: t = (low2 + high2) / 2; \
113: if (cp2[sliceheight * t] > col) high2 = t; \
114: else low2 = t; \
115: } \
116: for (_i = low2; _i < high2; _i++) { \
117: if (cp2[sliceheight * _i] > col) break; \
118: if (cp2[sliceheight * _i] == col) { \
119: if (addv == ADD_VALUES) vp2[sliceheight * _i] += value; \
120: else vp2[sliceheight * _i] = value; \
121: inserted = PETSC_TRUE; \
122: goto b_noinsert; \
123: } \
124: } \
125: if (value == 0.0 && ignorezeroentries) { \
126: low2 = 0; \
127: high2 = nrow2; \
128: goto b_noinsert; \
129: } \
130: if (nonew == 1) { \
131: low2 = 0; \
132: high2 = nrow2; \
133: goto b_noinsert; \
134: } \
135: PetscCheck(nonew != -1, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Inserting a new nonzero at global row/column (%" PetscInt_FMT ", %" PetscInt_FMT ") into matrix", orow, ocol); \
136: MatSeqXSELLReallocateSELL(B, bm, 1, nrow2, b->sliidx, b->sliceheight, row / sliceheight, row, col, b->colidx, b->val, cp2, vp2, nonew, MatScalar); \
137: /* shift up all the later entries in this row */ \
138: for (ii = nrow2 - 1; ii >= _i; ii--) { \
139: cp2[sliceheight * (ii + 1)] = cp2[sliceheight * ii]; \
140: vp2[sliceheight * (ii + 1)] = vp2[sliceheight * ii]; \
141: } \
142: cp2[sliceheight * _i] = col; \
143: vp2[sliceheight * _i] = value; \
144: b->nz++; \
145: nrow2++; \
146: b_noinsert:; \
147: b->rlen[row] = nrow2; \
148: }
150: static PetscErrorCode MatSetValues_MPISELL(Mat mat, PetscInt m, const PetscInt im[], PetscInt n, const PetscInt in[], const PetscScalar v[], InsertMode addv)
151: {
152: Mat_MPISELL *sell = (Mat_MPISELL *)mat->data;
153: PetscScalar value;
154: PetscInt i, j, rstart = mat->rmap->rstart, rend = mat->rmap->rend, shift1, shift2;
155: PetscInt cstart = mat->cmap->rstart, cend = mat->cmap->rend, row, col;
156: PetscBool roworiented = sell->roworiented;
158: /* Some Variables required in the macro */
159: Mat A = sell->A;
160: Mat_SeqSELL *a = (Mat_SeqSELL *)A->data;
161: PetscBool ignorezeroentries = a->ignorezeroentries, found;
162: Mat B = sell->B;
163: Mat_SeqSELL *b = (Mat_SeqSELL *)B->data;
164: PetscInt *cp1, *cp2, ii, _i, nrow1, nrow2, low1, high1, low2, high2, t, lastcol1, lastcol2, sliceheight = a->sliceheight;
165: MatScalar *vp1, *vp2;
167: PetscFunctionBegin;
168: for (i = 0; i < m; i++) {
169: if (im[i] < 0) continue;
170: PetscCheck(im[i] < mat->rmap->N, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Row too large: row %" PetscInt_FMT " max %" PetscInt_FMT, im[i], mat->rmap->N - 1);
171: if (im[i] >= rstart && im[i] < rend) {
172: row = im[i] - rstart;
173: lastcol1 = -1;
174: shift1 = a->sliidx[row / sliceheight] + (row % sliceheight); /* starting index of the row */
175: cp1 = PetscSafePointerPlusOffset(a->colidx, shift1);
176: vp1 = PetscSafePointerPlusOffset(a->val, shift1);
177: nrow1 = a->rlen[row];
178: low1 = 0;
179: high1 = nrow1;
180: lastcol2 = -1;
181: shift2 = b->sliidx[row / sliceheight] + (row % sliceheight); /* starting index of the row */
182: cp2 = PetscSafePointerPlusOffset(b->colidx, shift2);
183: vp2 = PetscSafePointerPlusOffset(b->val, shift2);
184: nrow2 = b->rlen[row];
185: low2 = 0;
186: high2 = nrow2;
188: for (j = 0; j < n; j++) {
189: if (roworiented) value = v[i * n + j];
190: else value = v[i + j * m];
191: if (ignorezeroentries && value == 0.0 && (addv == ADD_VALUES)) continue;
192: if (in[j] >= cstart && in[j] < cend) {
193: col = in[j] - cstart;
194: MatSetValue_SeqSELL_Private(A, row, col, value, addv, im[i], in[j], cp1, vp1, lastcol1, low1, high1); /* set one value */
195: #if defined(PETSC_HAVE_CUDA)
196: if (A->offloadmask != PETSC_OFFLOAD_UNALLOCATED && found) A->offloadmask = PETSC_OFFLOAD_CPU;
197: #endif
198: } else if (in[j] < 0) {
199: continue;
200: } else {
201: PetscCheck(in[j] < mat->cmap->N, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Column too large: col %" PetscInt_FMT " max %" PetscInt_FMT, in[j], mat->cmap->N - 1);
202: if (mat->was_assembled) {
203: if (!sell->colmap) PetscCall(MatCreateColmap_MPISELL_Private(mat));
204: #if defined(PETSC_USE_CTABLE)
205: PetscCall(PetscHMapIGetWithDefault(sell->colmap, in[j] + 1, 0, &col));
206: col--;
207: #else
208: col = sell->colmap[in[j]] - 1;
209: #endif
210: if (col < 0 && !((Mat_SeqSELL *)sell->B->data)->nonew) {
211: PetscCall(MatDisAssemble_MPISELL(mat));
212: col = in[j];
213: /* Reinitialize the variables required by MatSetValues_SeqSELL_B_Private() */
214: B = sell->B;
215: b = (Mat_SeqSELL *)B->data;
216: shift2 = b->sliidx[row / sliceheight] + (row % sliceheight); /* starting index of the row */
217: cp2 = b->colidx + shift2;
218: vp2 = b->val + shift2;
219: nrow2 = b->rlen[row];
220: low2 = 0;
221: high2 = nrow2;
222: found = PETSC_FALSE;
223: } else {
224: PetscCheck(col >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Inserting a new nonzero at global row/column (%" PetscInt_FMT ", %" PetscInt_FMT ") into matrix", im[i], in[j]);
225: }
226: } else col = in[j];
227: MatSetValue_SeqSELL_Private(B, row, col, value, addv, im[i], in[j], cp2, vp2, lastcol2, low2, high2); /* set one value */
228: #if defined(PETSC_HAVE_CUDA)
229: if (B->offloadmask != PETSC_OFFLOAD_UNALLOCATED && found) B->offloadmask = PETSC_OFFLOAD_CPU;
230: #endif
231: }
232: }
233: } else {
234: PetscCheck(!mat->nooffprocentries, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "Setting off process row %" PetscInt_FMT " even though MatSetOption(,MAT_NO_OFF_PROC_ENTRIES,PETSC_TRUE) was set", im[i]);
235: if (!sell->donotstash) {
236: mat->assembled = PETSC_FALSE;
237: if (roworiented) {
238: PetscCall(MatStashValuesRow_Private(&mat->stash, im[i], n, in, v + i * n, (PetscBool)(ignorezeroentries && (addv == ADD_VALUES))));
239: } else {
240: PetscCall(MatStashValuesCol_Private(&mat->stash, im[i], n, in, v + i, m, (PetscBool)(ignorezeroentries && (addv == ADD_VALUES))));
241: }
242: }
243: }
244: }
245: PetscFunctionReturn(PETSC_SUCCESS);
246: }
248: static PetscErrorCode MatGetValues_MPISELL(Mat mat, PetscInt m, const PetscInt idxm[], PetscInt n, const PetscInt idxn[], PetscScalar v[])
249: {
250: Mat_MPISELL *sell = (Mat_MPISELL *)mat->data;
251: PetscInt i, j, rstart = mat->rmap->rstart, rend = mat->rmap->rend;
252: PetscInt cstart = mat->cmap->rstart, cend = mat->cmap->rend, row, col;
254: PetscFunctionBegin;
255: for (i = 0; i < m; i++) {
256: if (idxm[i] < 0) continue; /* negative row */
257: PetscCheck(idxm[i] < mat->rmap->N, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Row too large: row %" PetscInt_FMT " max %" PetscInt_FMT, idxm[i], mat->rmap->N - 1);
258: if (idxm[i] >= rstart && idxm[i] < rend) {
259: row = idxm[i] - rstart;
260: for (j = 0; j < n; j++) {
261: if (idxn[j] < 0) continue; /* negative column */
262: PetscCheck(idxn[j] < mat->cmap->N, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Column too large: col %" PetscInt_FMT " max %" PetscInt_FMT, idxn[j], mat->cmap->N - 1);
263: if (idxn[j] >= cstart && idxn[j] < cend) {
264: col = idxn[j] - cstart;
265: PetscCall(MatGetValues(sell->A, 1, &row, 1, &col, v + i * n + j));
266: } else {
267: if (!sell->colmap) PetscCall(MatCreateColmap_MPISELL_Private(mat));
268: #if defined(PETSC_USE_CTABLE)
269: PetscCall(PetscHMapIGetWithDefault(sell->colmap, idxn[j] + 1, 0, &col));
270: col--;
271: #else
272: col = sell->colmap[idxn[j]] - 1;
273: #endif
274: if ((col < 0) || (sell->garray[col] != idxn[j])) *(v + i * n + j) = 0.0;
275: else PetscCall(MatGetValues(sell->B, 1, &row, 1, &col, v + i * n + j));
276: }
277: }
278: } else SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP, "Only local values currently supported");
279: }
280: PetscFunctionReturn(PETSC_SUCCESS);
281: }
283: static PetscErrorCode MatAssemblyBegin_MPISELL(Mat mat, MatAssemblyType mode)
284: {
285: Mat_MPISELL *sell = (Mat_MPISELL *)mat->data;
286: PetscInt nstash, reallocs;
288: PetscFunctionBegin;
289: if (sell->donotstash || mat->nooffprocentries) PetscFunctionReturn(PETSC_SUCCESS);
291: PetscCall(MatStashScatterBegin_Private(mat, &mat->stash, mat->rmap->range));
292: PetscCall(MatStashGetInfo_Private(&mat->stash, &nstash, &reallocs));
293: PetscCall(PetscInfo(sell->A, "Stash has %" PetscInt_FMT " entries, uses %" PetscInt_FMT " mallocs.\n", nstash, reallocs));
294: PetscFunctionReturn(PETSC_SUCCESS);
295: }
297: PetscErrorCode MatAssemblyEnd_MPISELL(Mat mat, MatAssemblyType mode)
298: {
299: Mat_MPISELL *sell = (Mat_MPISELL *)mat->data;
300: PetscMPIInt n;
301: PetscInt i, flg;
302: PetscInt *row, *col;
303: PetscScalar *val;
304: PetscBool all_assembled;
305: /* do not use 'b = (Mat_SeqSELL*)sell->B->data' as B can be reset in disassembly */
306: PetscFunctionBegin;
307: if (!sell->donotstash && !mat->nooffprocentries) {
308: while (1) {
309: PetscCall(MatStashScatterGetMesg_Private(&mat->stash, &n, &row, &col, &val, &flg));
310: if (!flg) break;
312: for (i = 0; i < n; i++) { /* assemble one by one */
313: PetscCall(MatSetValues_MPISELL(mat, 1, row + i, 1, col + i, val + i, mat->insertmode));
314: }
315: }
316: PetscCall(MatStashScatterEnd_Private(&mat->stash));
317: }
318: #if defined(PETSC_HAVE_CUDA)
319: if (mat->offloadmask == PETSC_OFFLOAD_CPU) sell->A->offloadmask = PETSC_OFFLOAD_CPU;
320: #endif
321: PetscCall(MatAssemblyBegin(sell->A, mode));
322: PetscCall(MatAssemblyEnd(sell->A, mode));
324: /*
325: determine if any process has disassembled, if so we must
326: also disassemble ourselves, in order that we may reassemble.
327: */
328: /*
329: if nonzero structure of submatrix B cannot change then we know that
330: no process disassembled thus we can skip this stuff
331: */
332: if (!((Mat_SeqSELL *)sell->B->data)->nonew) {
333: PetscCallMPI(MPIU_Allreduce(&mat->was_assembled, &all_assembled, 1, MPIU_BOOL, MPI_LAND, PetscObjectComm((PetscObject)mat)));
334: if (mat->was_assembled && !all_assembled) PetscCall(MatDisAssemble_MPISELL(mat));
335: }
336: if (!mat->was_assembled && mode == MAT_FINAL_ASSEMBLY) PetscCall(MatSetUpMultiply_MPISELL(mat));
337: #if defined(PETSC_HAVE_CUDA)
338: if (mat->offloadmask == PETSC_OFFLOAD_CPU && sell->B->offloadmask != PETSC_OFFLOAD_UNALLOCATED) sell->B->offloadmask = PETSC_OFFLOAD_CPU;
339: #endif
340: PetscCall(MatAssemblyBegin(sell->B, mode));
341: PetscCall(MatAssemblyEnd(sell->B, mode));
342: PetscCall(PetscFree2(sell->rowvalues, sell->rowindices));
343: sell->rowvalues = NULL;
344: PetscCall(VecDestroy(&sell->diag));
346: /* if no new nonzero locations are allowed in matrix then only set the matrix state the first time through */
347: if ((!mat->was_assembled && mode == MAT_FINAL_ASSEMBLY) || !((Mat_SeqSELL *)sell->A->data)->nonew) {
348: PetscObjectState state = sell->A->nonzerostate + sell->B->nonzerostate;
349: PetscCallMPI(MPIU_Allreduce(&state, &mat->nonzerostate, 1, MPIU_INT64, MPI_SUM, PetscObjectComm((PetscObject)mat)));
350: }
351: #if defined(PETSC_HAVE_CUDA)
352: mat->offloadmask = PETSC_OFFLOAD_BOTH;
353: #endif
354: PetscFunctionReturn(PETSC_SUCCESS);
355: }
357: static PetscErrorCode MatZeroEntries_MPISELL(Mat A)
358: {
359: Mat_MPISELL *l = (Mat_MPISELL *)A->data;
361: PetscFunctionBegin;
362: PetscCall(MatZeroEntries(l->A));
363: PetscCall(MatZeroEntries(l->B));
364: PetscFunctionReturn(PETSC_SUCCESS);
365: }
367: static PetscErrorCode MatMult_MPISELL(Mat A, Vec xx, Vec yy)
368: {
369: Mat_MPISELL *a = (Mat_MPISELL *)A->data;
370: PetscInt nt;
372: PetscFunctionBegin;
373: PetscCall(VecGetLocalSize(xx, &nt));
374: PetscCheck(nt == A->cmap->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Incompatible partition of A (%" PetscInt_FMT ") and xx (%" PetscInt_FMT ")", A->cmap->n, nt);
375: PetscCall(VecScatterBegin(a->Mvctx, xx, a->lvec, INSERT_VALUES, SCATTER_FORWARD));
376: PetscCall((*a->A->ops->mult)(a->A, xx, yy));
377: PetscCall(VecScatterEnd(a->Mvctx, xx, a->lvec, INSERT_VALUES, SCATTER_FORWARD));
378: PetscCall((*a->B->ops->multadd)(a->B, a->lvec, yy, yy));
379: PetscFunctionReturn(PETSC_SUCCESS);
380: }
382: static PetscErrorCode MatMultDiagonalBlock_MPISELL(Mat A, Vec bb, Vec xx)
383: {
384: Mat_MPISELL *a = (Mat_MPISELL *)A->data;
386: PetscFunctionBegin;
387: PetscCall(MatMultDiagonalBlock(a->A, bb, xx));
388: PetscFunctionReturn(PETSC_SUCCESS);
389: }
391: static PetscErrorCode MatMultAdd_MPISELL(Mat A, Vec xx, Vec yy, Vec zz)
392: {
393: Mat_MPISELL *a = (Mat_MPISELL *)A->data;
395: PetscFunctionBegin;
396: PetscCall(VecScatterBegin(a->Mvctx, xx, a->lvec, INSERT_VALUES, SCATTER_FORWARD));
397: PetscCall((*a->A->ops->multadd)(a->A, xx, yy, zz));
398: PetscCall(VecScatterEnd(a->Mvctx, xx, a->lvec, INSERT_VALUES, SCATTER_FORWARD));
399: PetscCall((*a->B->ops->multadd)(a->B, a->lvec, zz, zz));
400: PetscFunctionReturn(PETSC_SUCCESS);
401: }
403: static PetscErrorCode MatMultTranspose_MPISELL(Mat A, Vec xx, Vec yy)
404: {
405: Mat_MPISELL *a = (Mat_MPISELL *)A->data;
407: PetscFunctionBegin;
408: /* do nondiagonal part */
409: PetscCall((*a->B->ops->multtranspose)(a->B, xx, a->lvec));
410: /* do local part */
411: PetscCall((*a->A->ops->multtranspose)(a->A, xx, yy));
412: /* add partial results together */
413: PetscCall(VecScatterBegin(a->Mvctx, a->lvec, yy, ADD_VALUES, SCATTER_REVERSE));
414: PetscCall(VecScatterEnd(a->Mvctx, a->lvec, yy, ADD_VALUES, SCATTER_REVERSE));
415: PetscFunctionReturn(PETSC_SUCCESS);
416: }
418: static PetscErrorCode MatIsTranspose_MPISELL(Mat Amat, Mat Bmat, PetscReal tol, PetscBool *f)
419: {
420: MPI_Comm comm;
421: Mat_MPISELL *Asell = (Mat_MPISELL *)Amat->data, *Bsell;
422: Mat Adia = Asell->A, Bdia, Aoff, Boff, *Aoffs, *Boffs;
423: IS Me, Notme;
424: PetscInt M, N, first, last, *notme, i;
425: PetscMPIInt size;
427: PetscFunctionBegin;
428: /* Easy test: symmetric diagonal block */
429: Bsell = (Mat_MPISELL *)Bmat->data;
430: Bdia = Bsell->A;
431: PetscCall(MatIsTranspose(Adia, Bdia, tol, f));
432: if (!*f) PetscFunctionReturn(PETSC_SUCCESS);
433: PetscCall(PetscObjectGetComm((PetscObject)Amat, &comm));
434: PetscCallMPI(MPI_Comm_size(comm, &size));
435: if (size == 1) PetscFunctionReturn(PETSC_SUCCESS);
437: /* Hard test: off-diagonal block. This takes a MatCreateSubMatrix. */
438: PetscCall(MatGetSize(Amat, &M, &N));
439: PetscCall(MatGetOwnershipRange(Amat, &first, &last));
440: PetscCall(PetscMalloc1(N - last + first, ¬me));
441: for (i = 0; i < first; i++) notme[i] = i;
442: for (i = last; i < M; i++) notme[i - last + first] = i;
443: PetscCall(ISCreateGeneral(MPI_COMM_SELF, N - last + first, notme, PETSC_COPY_VALUES, &Notme));
444: PetscCall(ISCreateStride(MPI_COMM_SELF, last - first, first, 1, &Me));
445: PetscCall(MatCreateSubMatrices(Amat, 1, &Me, &Notme, MAT_INITIAL_MATRIX, &Aoffs));
446: Aoff = Aoffs[0];
447: PetscCall(MatCreateSubMatrices(Bmat, 1, &Notme, &Me, MAT_INITIAL_MATRIX, &Boffs));
448: Boff = Boffs[0];
449: PetscCall(MatIsTranspose(Aoff, Boff, tol, f));
450: PetscCall(MatDestroyMatrices(1, &Aoffs));
451: PetscCall(MatDestroyMatrices(1, &Boffs));
452: PetscCall(ISDestroy(&Me));
453: PetscCall(ISDestroy(&Notme));
454: PetscCall(PetscFree(notme));
455: PetscFunctionReturn(PETSC_SUCCESS);
456: }
458: static PetscErrorCode MatMultTransposeAdd_MPISELL(Mat A, Vec xx, Vec yy, Vec zz)
459: {
460: Mat_MPISELL *a = (Mat_MPISELL *)A->data;
462: PetscFunctionBegin;
463: /* do nondiagonal part */
464: PetscCall((*a->B->ops->multtranspose)(a->B, xx, a->lvec));
465: /* do local part */
466: PetscCall((*a->A->ops->multtransposeadd)(a->A, xx, yy, zz));
467: /* add partial results together */
468: PetscCall(VecScatterBegin(a->Mvctx, a->lvec, zz, ADD_VALUES, SCATTER_REVERSE));
469: PetscCall(VecScatterEnd(a->Mvctx, a->lvec, zz, ADD_VALUES, SCATTER_REVERSE));
470: PetscFunctionReturn(PETSC_SUCCESS);
471: }
473: /*
474: This only works correctly for square matrices where the subblock A->A is the
475: diagonal block
476: */
477: static PetscErrorCode MatGetDiagonal_MPISELL(Mat A, Vec v)
478: {
479: Mat_MPISELL *a = (Mat_MPISELL *)A->data;
481: PetscFunctionBegin;
482: PetscCheck(A->rmap->N == A->cmap->N, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Supports only square matrix where A->A is diag block");
483: PetscCheck(A->rmap->rstart == A->cmap->rstart && A->rmap->rend == A->cmap->rend, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "row partition must equal col partition");
484: PetscCall(MatGetDiagonal(a->A, v));
485: PetscFunctionReturn(PETSC_SUCCESS);
486: }
488: static PetscErrorCode MatScale_MPISELL(Mat A, PetscScalar aa)
489: {
490: Mat_MPISELL *a = (Mat_MPISELL *)A->data;
492: PetscFunctionBegin;
493: PetscCall(MatScale(a->A, aa));
494: PetscCall(MatScale(a->B, aa));
495: PetscFunctionReturn(PETSC_SUCCESS);
496: }
498: PetscErrorCode MatDestroy_MPISELL(Mat mat)
499: {
500: Mat_MPISELL *sell = (Mat_MPISELL *)mat->data;
502: PetscFunctionBegin;
503: PetscCall(PetscLogObjectState((PetscObject)mat, "Rows=%" PetscInt_FMT ", Cols=%" PetscInt_FMT, mat->rmap->N, mat->cmap->N));
504: PetscCall(MatStashDestroy_Private(&mat->stash));
505: PetscCall(VecDestroy(&sell->diag));
506: PetscCall(MatDestroy(&sell->A));
507: PetscCall(MatDestroy(&sell->B));
508: #if defined(PETSC_USE_CTABLE)
509: PetscCall(PetscHMapIDestroy(&sell->colmap));
510: #else
511: PetscCall(PetscFree(sell->colmap));
512: #endif
513: PetscCall(PetscFree(sell->garray));
514: PetscCall(VecDestroy(&sell->lvec));
515: PetscCall(VecScatterDestroy(&sell->Mvctx));
516: PetscCall(PetscFree2(sell->rowvalues, sell->rowindices));
517: PetscCall(PetscFree(sell->ld));
518: PetscCall(PetscFree(mat->data));
520: PetscCall(PetscObjectChangeTypeName((PetscObject)mat, NULL));
521: PetscCall(PetscObjectComposeFunction((PetscObject)mat, "MatStoreValues_C", NULL));
522: PetscCall(PetscObjectComposeFunction((PetscObject)mat, "MatRetrieveValues_C", NULL));
523: PetscCall(PetscObjectComposeFunction((PetscObject)mat, "MatIsTranspose_C", NULL));
524: PetscCall(PetscObjectComposeFunction((PetscObject)mat, "MatMPISELLSetPreallocation_C", NULL));
525: PetscCall(PetscObjectComposeFunction((PetscObject)mat, "MatConvert_mpisell_mpiaij_C", NULL));
526: #if defined(PETSC_HAVE_CUDA)
527: PetscCall(PetscObjectComposeFunction((PetscObject)mat, "MatConvert_mpisell_mpisellcuda_C", NULL));
528: #endif
529: PetscCall(PetscObjectComposeFunction((PetscObject)mat, "MatDiagonalScaleLocal_C", NULL));
530: PetscFunctionReturn(PETSC_SUCCESS);
531: }
533: #include <petscdraw.h>
534: static PetscErrorCode MatView_MPISELL_ASCIIorDraworSocket(Mat mat, PetscViewer viewer)
535: {
536: Mat_MPISELL *sell = (Mat_MPISELL *)mat->data;
537: PetscMPIInt rank = sell->rank, size = sell->size;
538: PetscBool isdraw, iascii, isbinary;
539: PetscViewer sviewer;
540: PetscViewerFormat format;
542: PetscFunctionBegin;
543: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERDRAW, &isdraw));
544: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &iascii));
545: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERBINARY, &isbinary));
546: if (iascii) {
547: PetscCall(PetscViewerGetFormat(viewer, &format));
548: if (format == PETSC_VIEWER_ASCII_INFO_DETAIL) {
549: MatInfo info;
550: PetscInt *inodes;
552: PetscCallMPI(MPI_Comm_rank(PetscObjectComm((PetscObject)mat), &rank));
553: PetscCall(MatGetInfo(mat, MAT_LOCAL, &info));
554: PetscCall(MatInodeGetInodeSizes(sell->A, NULL, &inodes, NULL));
555: PetscCall(PetscViewerASCIIPushSynchronized(viewer));
556: if (!inodes) {
557: PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "[%d] Local rows %" PetscInt_FMT " nz %" PetscInt_FMT " nz alloced %" PetscInt_FMT " mem %" PetscInt_FMT ", not using I-node routines\n", rank, mat->rmap->n, (PetscInt)info.nz_used,
558: (PetscInt)info.nz_allocated, (PetscInt)info.memory));
559: } else {
560: PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "[%d] Local rows %" PetscInt_FMT " nz %" PetscInt_FMT " nz alloced %" PetscInt_FMT " mem %" PetscInt_FMT ", using I-node routines\n", rank, mat->rmap->n, (PetscInt)info.nz_used,
561: (PetscInt)info.nz_allocated, (PetscInt)info.memory));
562: }
563: PetscCall(MatGetInfo(sell->A, MAT_LOCAL, &info));
564: PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "[%d] on-diagonal part: nz %" PetscInt_FMT " \n", rank, (PetscInt)info.nz_used));
565: PetscCall(MatGetInfo(sell->B, MAT_LOCAL, &info));
566: PetscCall(PetscViewerASCIISynchronizedPrintf(viewer, "[%d] off-diagonal part: nz %" PetscInt_FMT " \n", rank, (PetscInt)info.nz_used));
567: PetscCall(PetscViewerFlush(viewer));
568: PetscCall(PetscViewerASCIIPopSynchronized(viewer));
569: PetscCall(PetscViewerASCIIPrintf(viewer, "Information on VecScatter used in matrix-vector product: \n"));
570: PetscCall(VecScatterView(sell->Mvctx, viewer));
571: PetscFunctionReturn(PETSC_SUCCESS);
572: } else if (format == PETSC_VIEWER_ASCII_INFO) {
573: PetscInt inodecount, inodelimit, *inodes;
574: PetscCall(MatInodeGetInodeSizes(sell->A, &inodecount, &inodes, &inodelimit));
575: if (inodes) {
576: PetscCall(PetscViewerASCIIPrintf(viewer, "using I-node (on process 0) routines: found %" PetscInt_FMT " nodes, limit used is %" PetscInt_FMT "\n", inodecount, inodelimit));
577: } else {
578: PetscCall(PetscViewerASCIIPrintf(viewer, "not using I-node (on process 0) routines\n"));
579: }
580: PetscFunctionReturn(PETSC_SUCCESS);
581: } else if (format == PETSC_VIEWER_ASCII_FACTOR_INFO) {
582: PetscFunctionReturn(PETSC_SUCCESS);
583: }
584: } else if (isbinary) {
585: if (size == 1) {
586: PetscCall(PetscObjectSetName((PetscObject)sell->A, ((PetscObject)mat)->name));
587: PetscCall(MatView(sell->A, viewer));
588: } else {
589: /* PetscCall(MatView_MPISELL_Binary(mat,viewer)); */
590: }
591: PetscFunctionReturn(PETSC_SUCCESS);
592: } else if (isdraw) {
593: PetscDraw draw;
594: PetscBool isnull;
595: PetscCall(PetscViewerDrawGetDraw(viewer, 0, &draw));
596: PetscCall(PetscDrawIsNull(draw, &isnull));
597: if (isnull) PetscFunctionReturn(PETSC_SUCCESS);
598: }
600: {
601: /* assemble the entire matrix onto first processor. */
602: Mat A;
603: Mat_SeqSELL *Aloc;
604: PetscInt M = mat->rmap->N, N = mat->cmap->N, *acolidx, row, col, i, j;
605: MatScalar *aval;
606: PetscBool isnonzero;
608: PetscCall(MatCreate(PetscObjectComm((PetscObject)mat), &A));
609: if (rank == 0) {
610: PetscCall(MatSetSizes(A, M, N, M, N));
611: } else {
612: PetscCall(MatSetSizes(A, 0, 0, M, N));
613: }
614: /* This is just a temporary matrix, so explicitly using MATMPISELL is probably best */
615: PetscCall(MatSetType(A, MATMPISELL));
616: PetscCall(MatMPISELLSetPreallocation(A, 0, NULL, 0, NULL));
617: PetscCall(MatSetOption(A, MAT_NEW_NONZERO_LOCATION_ERR, PETSC_FALSE));
619: /* copy over the A part */
620: Aloc = (Mat_SeqSELL *)sell->A->data;
621: acolidx = Aloc->colidx;
622: aval = Aloc->val;
623: for (i = 0; i < Aloc->totalslices; i++) { /* loop over slices */
624: for (j = Aloc->sliidx[i]; j < Aloc->sliidx[i + 1]; j++) {
625: isnonzero = (PetscBool)((j - Aloc->sliidx[i]) / Aloc->sliceheight < Aloc->rlen[i * Aloc->sliceheight + j % Aloc->sliceheight]);
626: if (isnonzero) { /* check the mask bit */
627: row = i * Aloc->sliceheight + j % Aloc->sliceheight + mat->rmap->rstart;
628: col = *acolidx + mat->rmap->rstart;
629: PetscCall(MatSetValues(A, 1, &row, 1, &col, aval, INSERT_VALUES));
630: }
631: aval++;
632: acolidx++;
633: }
634: }
636: /* copy over the B part */
637: Aloc = (Mat_SeqSELL *)sell->B->data;
638: acolidx = Aloc->colidx;
639: aval = Aloc->val;
640: for (i = 0; i < Aloc->totalslices; i++) {
641: for (j = Aloc->sliidx[i]; j < Aloc->sliidx[i + 1]; j++) {
642: isnonzero = (PetscBool)((j - Aloc->sliidx[i]) / Aloc->sliceheight < Aloc->rlen[i * Aloc->sliceheight + j % Aloc->sliceheight]);
643: if (isnonzero) {
644: row = i * Aloc->sliceheight + j % Aloc->sliceheight + mat->rmap->rstart;
645: col = sell->garray[*acolidx];
646: PetscCall(MatSetValues(A, 1, &row, 1, &col, aval, INSERT_VALUES));
647: }
648: aval++;
649: acolidx++;
650: }
651: }
653: PetscCall(MatAssemblyBegin(A, MAT_FINAL_ASSEMBLY));
654: PetscCall(MatAssemblyEnd(A, MAT_FINAL_ASSEMBLY));
655: /*
656: Everyone has to call to draw the matrix since the graphics waits are
657: synchronized across all processors that share the PetscDraw object
658: */
659: PetscCall(PetscViewerGetSubViewer(viewer, PETSC_COMM_SELF, &sviewer));
660: if (rank == 0) {
661: PetscCall(PetscObjectSetName((PetscObject)((Mat_MPISELL *)A->data)->A, ((PetscObject)mat)->name));
662: PetscCall(MatView_SeqSELL(((Mat_MPISELL *)A->data)->A, sviewer));
663: }
664: PetscCall(PetscViewerRestoreSubViewer(viewer, PETSC_COMM_SELF, &sviewer));
665: PetscCall(MatDestroy(&A));
666: }
667: PetscFunctionReturn(PETSC_SUCCESS);
668: }
670: static PetscErrorCode MatView_MPISELL(Mat mat, PetscViewer viewer)
671: {
672: PetscBool iascii, isdraw, issocket, isbinary;
674: PetscFunctionBegin;
675: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &iascii));
676: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERDRAW, &isdraw));
677: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERBINARY, &isbinary));
678: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERSOCKET, &issocket));
679: if (iascii || isdraw || isbinary || issocket) PetscCall(MatView_MPISELL_ASCIIorDraworSocket(mat, viewer));
680: PetscFunctionReturn(PETSC_SUCCESS);
681: }
683: static PetscErrorCode MatGetGhosts_MPISELL(Mat mat, PetscInt *nghosts, const PetscInt *ghosts[])
684: {
685: Mat_MPISELL *sell = (Mat_MPISELL *)mat->data;
687: PetscFunctionBegin;
688: PetscCall(MatGetSize(sell->B, NULL, nghosts));
689: if (ghosts) *ghosts = sell->garray;
690: PetscFunctionReturn(PETSC_SUCCESS);
691: }
693: static PetscErrorCode MatGetInfo_MPISELL(Mat matin, MatInfoType flag, MatInfo *info)
694: {
695: Mat_MPISELL *mat = (Mat_MPISELL *)matin->data;
696: Mat A = mat->A, B = mat->B;
697: PetscLogDouble isend[5], irecv[5];
699: PetscFunctionBegin;
700: info->block_size = 1.0;
701: PetscCall(MatGetInfo(A, MAT_LOCAL, info));
703: isend[0] = info->nz_used;
704: isend[1] = info->nz_allocated;
705: isend[2] = info->nz_unneeded;
706: isend[3] = info->memory;
707: isend[4] = info->mallocs;
709: PetscCall(MatGetInfo(B, MAT_LOCAL, info));
711: isend[0] += info->nz_used;
712: isend[1] += info->nz_allocated;
713: isend[2] += info->nz_unneeded;
714: isend[3] += info->memory;
715: isend[4] += info->mallocs;
716: if (flag == MAT_LOCAL) {
717: info->nz_used = isend[0];
718: info->nz_allocated = isend[1];
719: info->nz_unneeded = isend[2];
720: info->memory = isend[3];
721: info->mallocs = isend[4];
722: } else if (flag == MAT_GLOBAL_MAX) {
723: PetscCallMPI(MPIU_Allreduce(isend, irecv, 5, MPIU_PETSCLOGDOUBLE, MPI_MAX, PetscObjectComm((PetscObject)matin)));
725: info->nz_used = irecv[0];
726: info->nz_allocated = irecv[1];
727: info->nz_unneeded = irecv[2];
728: info->memory = irecv[3];
729: info->mallocs = irecv[4];
730: } else if (flag == MAT_GLOBAL_SUM) {
731: PetscCallMPI(MPIU_Allreduce(isend, irecv, 5, MPIU_PETSCLOGDOUBLE, MPI_SUM, PetscObjectComm((PetscObject)matin)));
733: info->nz_used = irecv[0];
734: info->nz_allocated = irecv[1];
735: info->nz_unneeded = irecv[2];
736: info->memory = irecv[3];
737: info->mallocs = irecv[4];
738: }
739: info->fill_ratio_given = 0; /* no parallel LU/ILU/Cholesky */
740: info->fill_ratio_needed = 0;
741: info->factor_mallocs = 0;
742: PetscFunctionReturn(PETSC_SUCCESS);
743: }
745: static PetscErrorCode MatSetOption_MPISELL(Mat A, MatOption op, PetscBool flg)
746: {
747: Mat_MPISELL *a = (Mat_MPISELL *)A->data;
749: PetscFunctionBegin;
750: switch (op) {
751: case MAT_NEW_NONZERO_LOCATIONS:
752: case MAT_NEW_NONZERO_ALLOCATION_ERR:
753: case MAT_UNUSED_NONZERO_LOCATION_ERR:
754: case MAT_KEEP_NONZERO_PATTERN:
755: case MAT_NEW_NONZERO_LOCATION_ERR:
756: case MAT_USE_INODES:
757: case MAT_IGNORE_ZERO_ENTRIES:
758: MatCheckPreallocated(A, 1);
759: PetscCall(MatSetOption(a->A, op, flg));
760: PetscCall(MatSetOption(a->B, op, flg));
761: break;
762: case MAT_ROW_ORIENTED:
763: MatCheckPreallocated(A, 1);
764: a->roworiented = flg;
766: PetscCall(MatSetOption(a->A, op, flg));
767: PetscCall(MatSetOption(a->B, op, flg));
768: break;
769: case MAT_IGNORE_OFF_PROC_ENTRIES:
770: a->donotstash = flg;
771: break;
772: case MAT_SYMMETRIC:
773: MatCheckPreallocated(A, 1);
774: PetscCall(MatSetOption(a->A, op, flg));
775: break;
776: case MAT_STRUCTURALLY_SYMMETRIC:
777: MatCheckPreallocated(A, 1);
778: PetscCall(MatSetOption(a->A, op, flg));
779: break;
780: case MAT_HERMITIAN:
781: MatCheckPreallocated(A, 1);
782: PetscCall(MatSetOption(a->A, op, flg));
783: break;
784: case MAT_SYMMETRY_ETERNAL:
785: MatCheckPreallocated(A, 1);
786: PetscCall(MatSetOption(a->A, op, flg));
787: break;
788: case MAT_STRUCTURAL_SYMMETRY_ETERNAL:
789: MatCheckPreallocated(A, 1);
790: PetscCall(MatSetOption(a->A, op, flg));
791: break;
792: default:
793: break;
794: }
795: PetscFunctionReturn(PETSC_SUCCESS);
796: }
798: static PetscErrorCode MatDiagonalScale_MPISELL(Mat mat, Vec ll, Vec rr)
799: {
800: Mat_MPISELL *sell = (Mat_MPISELL *)mat->data;
801: Mat a = sell->A, b = sell->B;
802: PetscInt s1, s2, s3;
804: PetscFunctionBegin;
805: PetscCall(MatGetLocalSize(mat, &s2, &s3));
806: if (rr) {
807: PetscCall(VecGetLocalSize(rr, &s1));
808: PetscCheck(s1 == s3, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "right vector non-conforming local size");
809: /* Overlap communication with computation. */
810: PetscCall(VecScatterBegin(sell->Mvctx, rr, sell->lvec, INSERT_VALUES, SCATTER_FORWARD));
811: }
812: if (ll) {
813: PetscCall(VecGetLocalSize(ll, &s1));
814: PetscCheck(s1 == s2, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "left vector non-conforming local size");
815: PetscUseTypeMethod(b, diagonalscale, ll, NULL);
816: }
817: /* scale the diagonal block */
818: PetscUseTypeMethod(a, diagonalscale, ll, rr);
820: if (rr) {
821: /* Do a scatter end and then right scale the off-diagonal block */
822: PetscCall(VecScatterEnd(sell->Mvctx, rr, sell->lvec, INSERT_VALUES, SCATTER_FORWARD));
823: PetscUseTypeMethod(b, diagonalscale, NULL, sell->lvec);
824: }
825: PetscFunctionReturn(PETSC_SUCCESS);
826: }
828: static PetscErrorCode MatSetUnfactored_MPISELL(Mat A)
829: {
830: Mat_MPISELL *a = (Mat_MPISELL *)A->data;
832: PetscFunctionBegin;
833: PetscCall(MatSetUnfactored(a->A));
834: PetscFunctionReturn(PETSC_SUCCESS);
835: }
837: static PetscErrorCode MatEqual_MPISELL(Mat A, Mat B, PetscBool *flag)
838: {
839: Mat_MPISELL *matB = (Mat_MPISELL *)B->data, *matA = (Mat_MPISELL *)A->data;
840: Mat a, b, c, d;
841: PetscBool flg;
843: PetscFunctionBegin;
844: a = matA->A;
845: b = matA->B;
846: c = matB->A;
847: d = matB->B;
849: PetscCall(MatEqual(a, c, &flg));
850: if (flg) PetscCall(MatEqual(b, d, &flg));
851: PetscCallMPI(MPIU_Allreduce(&flg, flag, 1, MPIU_BOOL, MPI_LAND, PetscObjectComm((PetscObject)A)));
852: PetscFunctionReturn(PETSC_SUCCESS);
853: }
855: static PetscErrorCode MatCopy_MPISELL(Mat A, Mat B, MatStructure str)
856: {
857: Mat_MPISELL *a = (Mat_MPISELL *)A->data;
858: Mat_MPISELL *b = (Mat_MPISELL *)B->data;
860: PetscFunctionBegin;
861: /* If the two matrices don't have the same copy implementation, they aren't compatible for fast copy. */
862: if ((str != SAME_NONZERO_PATTERN) || (A->ops->copy != B->ops->copy)) {
863: /* because of the column compression in the off-processor part of the matrix a->B,
864: the number of columns in a->B and b->B may be different, hence we cannot call
865: the MatCopy() directly on the two parts. If need be, we can provide a more
866: efficient copy than the MatCopy_Basic() by first uncompressing the a->B matrices
867: then copying the submatrices */
868: PetscCall(MatCopy_Basic(A, B, str));
869: } else {
870: PetscCall(MatCopy(a->A, b->A, str));
871: PetscCall(MatCopy(a->B, b->B, str));
872: }
873: PetscFunctionReturn(PETSC_SUCCESS);
874: }
876: static PetscErrorCode MatSetUp_MPISELL(Mat A)
877: {
878: PetscFunctionBegin;
879: PetscCall(MatMPISELLSetPreallocation(A, PETSC_DEFAULT, NULL, PETSC_DEFAULT, NULL));
880: PetscFunctionReturn(PETSC_SUCCESS);
881: }
883: static PetscErrorCode MatConjugate_MPISELL(Mat mat)
884: {
885: PetscFunctionBegin;
886: if (PetscDefined(USE_COMPLEX)) {
887: Mat_MPISELL *sell = (Mat_MPISELL *)mat->data;
889: PetscCall(MatConjugate_SeqSELL(sell->A));
890: PetscCall(MatConjugate_SeqSELL(sell->B));
891: }
892: PetscFunctionReturn(PETSC_SUCCESS);
893: }
895: static PetscErrorCode MatRealPart_MPISELL(Mat A)
896: {
897: Mat_MPISELL *a = (Mat_MPISELL *)A->data;
899: PetscFunctionBegin;
900: PetscCall(MatRealPart(a->A));
901: PetscCall(MatRealPart(a->B));
902: PetscFunctionReturn(PETSC_SUCCESS);
903: }
905: static PetscErrorCode MatImaginaryPart_MPISELL(Mat A)
906: {
907: Mat_MPISELL *a = (Mat_MPISELL *)A->data;
909: PetscFunctionBegin;
910: PetscCall(MatImaginaryPart(a->A));
911: PetscCall(MatImaginaryPart(a->B));
912: PetscFunctionReturn(PETSC_SUCCESS);
913: }
915: static PetscErrorCode MatInvertBlockDiagonal_MPISELL(Mat A, const PetscScalar **values)
916: {
917: Mat_MPISELL *a = (Mat_MPISELL *)A->data;
919: PetscFunctionBegin;
920: PetscCall(MatInvertBlockDiagonal(a->A, values));
921: A->factorerrortype = a->A->factorerrortype;
922: PetscFunctionReturn(PETSC_SUCCESS);
923: }
925: static PetscErrorCode MatSetRandom_MPISELL(Mat x, PetscRandom rctx)
926: {
927: Mat_MPISELL *sell = (Mat_MPISELL *)x->data;
929: PetscFunctionBegin;
930: PetscCall(MatSetRandom(sell->A, rctx));
931: PetscCall(MatSetRandom(sell->B, rctx));
932: PetscCall(MatAssemblyBegin(x, MAT_FINAL_ASSEMBLY));
933: PetscCall(MatAssemblyEnd(x, MAT_FINAL_ASSEMBLY));
934: PetscFunctionReturn(PETSC_SUCCESS);
935: }
937: static PetscErrorCode MatSetFromOptions_MPISELL(Mat A, PetscOptionItems PetscOptionsObject)
938: {
939: PetscFunctionBegin;
940: PetscOptionsHeadBegin(PetscOptionsObject, "MPISELL options");
941: PetscOptionsHeadEnd();
942: PetscFunctionReturn(PETSC_SUCCESS);
943: }
945: static PetscErrorCode MatShift_MPISELL(Mat Y, PetscScalar a)
946: {
947: Mat_MPISELL *msell = (Mat_MPISELL *)Y->data;
948: Mat_SeqSELL *sell = (Mat_SeqSELL *)msell->A->data;
950: PetscFunctionBegin;
951: if (!Y->preallocated) {
952: PetscCall(MatMPISELLSetPreallocation(Y, 1, NULL, 0, NULL));
953: } else if (!sell->nz) {
954: PetscInt nonew = sell->nonew;
955: PetscCall(MatSeqSELLSetPreallocation(msell->A, 1, NULL));
956: sell->nonew = nonew;
957: }
958: PetscCall(MatShift_Basic(Y, a));
959: PetscFunctionReturn(PETSC_SUCCESS);
960: }
962: static PetscErrorCode MatMissingDiagonal_MPISELL(Mat A, PetscBool *missing, PetscInt *d)
963: {
964: Mat_MPISELL *a = (Mat_MPISELL *)A->data;
966: PetscFunctionBegin;
967: PetscCheck(A->rmap->n == A->cmap->n, PETSC_COMM_SELF, PETSC_ERR_SUP, "Only works for square matrices");
968: PetscCall(MatMissingDiagonal(a->A, missing, d));
969: if (d) {
970: PetscInt rstart;
971: PetscCall(MatGetOwnershipRange(A, &rstart, NULL));
972: *d += rstart;
973: }
974: PetscFunctionReturn(PETSC_SUCCESS);
975: }
977: static PetscErrorCode MatGetDiagonalBlock_MPISELL(Mat A, Mat *a)
978: {
979: PetscFunctionBegin;
980: *a = ((Mat_MPISELL *)A->data)->A;
981: PetscFunctionReturn(PETSC_SUCCESS);
982: }
984: static PetscErrorCode MatStoreValues_MPISELL(Mat mat)
985: {
986: Mat_MPISELL *sell = (Mat_MPISELL *)mat->data;
988: PetscFunctionBegin;
989: PetscCall(MatStoreValues(sell->A));
990: PetscCall(MatStoreValues(sell->B));
991: PetscFunctionReturn(PETSC_SUCCESS);
992: }
994: static PetscErrorCode MatRetrieveValues_MPISELL(Mat mat)
995: {
996: Mat_MPISELL *sell = (Mat_MPISELL *)mat->data;
998: PetscFunctionBegin;
999: PetscCall(MatRetrieveValues(sell->A));
1000: PetscCall(MatRetrieveValues(sell->B));
1001: PetscFunctionReturn(PETSC_SUCCESS);
1002: }
1004: static PetscErrorCode MatMPISELLSetPreallocation_MPISELL(Mat B, PetscInt d_rlenmax, const PetscInt d_rlen[], PetscInt o_rlenmax, const PetscInt o_rlen[])
1005: {
1006: Mat_MPISELL *b;
1008: PetscFunctionBegin;
1009: PetscCall(PetscLayoutSetUp(B->rmap));
1010: PetscCall(PetscLayoutSetUp(B->cmap));
1011: b = (Mat_MPISELL *)B->data;
1013: if (!B->preallocated) {
1014: /* Explicitly create 2 MATSEQSELL matrices. */
1015: PetscCall(MatCreate(PETSC_COMM_SELF, &b->A));
1016: PetscCall(MatSetSizes(b->A, B->rmap->n, B->cmap->n, B->rmap->n, B->cmap->n));
1017: PetscCall(MatSetBlockSizesFromMats(b->A, B, B));
1018: PetscCall(MatSetType(b->A, MATSEQSELL));
1019: PetscCall(MatCreate(PETSC_COMM_SELF, &b->B));
1020: PetscCall(MatSetSizes(b->B, B->rmap->n, B->cmap->N, B->rmap->n, B->cmap->N));
1021: PetscCall(MatSetBlockSizesFromMats(b->B, B, B));
1022: PetscCall(MatSetType(b->B, MATSEQSELL));
1023: }
1025: PetscCall(MatSeqSELLSetPreallocation(b->A, d_rlenmax, d_rlen));
1026: PetscCall(MatSeqSELLSetPreallocation(b->B, o_rlenmax, o_rlen));
1027: B->preallocated = PETSC_TRUE;
1028: B->was_assembled = PETSC_FALSE;
1029: /*
1030: critical for MatAssemblyEnd to work.
1031: MatAssemblyBegin checks it to set up was_assembled
1032: and MatAssemblyEnd checks was_assembled to determine whether to build garray
1033: */
1034: B->assembled = PETSC_FALSE;
1035: PetscFunctionReturn(PETSC_SUCCESS);
1036: }
1038: static PetscErrorCode MatDuplicate_MPISELL(Mat matin, MatDuplicateOption cpvalues, Mat *newmat)
1039: {
1040: Mat mat;
1041: Mat_MPISELL *a, *oldmat = (Mat_MPISELL *)matin->data;
1043: PetscFunctionBegin;
1044: *newmat = NULL;
1045: PetscCall(MatCreate(PetscObjectComm((PetscObject)matin), &mat));
1046: PetscCall(MatSetSizes(mat, matin->rmap->n, matin->cmap->n, matin->rmap->N, matin->cmap->N));
1047: PetscCall(MatSetBlockSizesFromMats(mat, matin, matin));
1048: PetscCall(MatSetType(mat, ((PetscObject)matin)->type_name));
1049: a = (Mat_MPISELL *)mat->data;
1051: mat->factortype = matin->factortype;
1052: mat->assembled = PETSC_TRUE;
1053: mat->insertmode = NOT_SET_VALUES;
1054: mat->preallocated = PETSC_TRUE;
1056: a->size = oldmat->size;
1057: a->rank = oldmat->rank;
1058: a->donotstash = oldmat->donotstash;
1059: a->roworiented = oldmat->roworiented;
1060: a->rowindices = NULL;
1061: a->rowvalues = NULL;
1062: a->getrowactive = PETSC_FALSE;
1064: PetscCall(PetscLayoutReference(matin->rmap, &mat->rmap));
1065: PetscCall(PetscLayoutReference(matin->cmap, &mat->cmap));
1067: if (oldmat->colmap) {
1068: #if defined(PETSC_USE_CTABLE)
1069: PetscCall(PetscHMapIDuplicate(oldmat->colmap, &a->colmap));
1070: #else
1071: PetscCall(PetscMalloc1(mat->cmap->N, &a->colmap));
1072: PetscCall(PetscArraycpy(a->colmap, oldmat->colmap, mat->cmap->N));
1073: #endif
1074: } else a->colmap = NULL;
1075: if (oldmat->garray) {
1076: PetscInt len;
1077: len = oldmat->B->cmap->n;
1078: PetscCall(PetscMalloc1(len + 1, &a->garray));
1079: if (len) PetscCall(PetscArraycpy(a->garray, oldmat->garray, len));
1080: } else a->garray = NULL;
1082: PetscCall(VecDuplicate(oldmat->lvec, &a->lvec));
1083: PetscCall(VecScatterCopy(oldmat->Mvctx, &a->Mvctx));
1084: PetscCall(MatDuplicate(oldmat->A, cpvalues, &a->A));
1085: PetscCall(MatDuplicate(oldmat->B, cpvalues, &a->B));
1086: PetscCall(PetscFunctionListDuplicate(((PetscObject)matin)->qlist, &((PetscObject)mat)->qlist));
1087: *newmat = mat;
1088: PetscFunctionReturn(PETSC_SUCCESS);
1089: }
1091: static const struct _MatOps MatOps_Values = {MatSetValues_MPISELL,
1092: NULL,
1093: NULL,
1094: MatMult_MPISELL,
1095: /* 4*/ MatMultAdd_MPISELL,
1096: MatMultTranspose_MPISELL,
1097: MatMultTransposeAdd_MPISELL,
1098: NULL,
1099: NULL,
1100: NULL,
1101: /*10*/ NULL,
1102: NULL,
1103: NULL,
1104: MatSOR_MPISELL,
1105: NULL,
1106: /*15*/ MatGetInfo_MPISELL,
1107: MatEqual_MPISELL,
1108: MatGetDiagonal_MPISELL,
1109: MatDiagonalScale_MPISELL,
1110: NULL,
1111: /*20*/ MatAssemblyBegin_MPISELL,
1112: MatAssemblyEnd_MPISELL,
1113: MatSetOption_MPISELL,
1114: MatZeroEntries_MPISELL,
1115: /*24*/ NULL,
1116: NULL,
1117: NULL,
1118: NULL,
1119: NULL,
1120: /*29*/ MatSetUp_MPISELL,
1121: NULL,
1122: NULL,
1123: MatGetDiagonalBlock_MPISELL,
1124: NULL,
1125: /*34*/ MatDuplicate_MPISELL,
1126: NULL,
1127: NULL,
1128: NULL,
1129: NULL,
1130: /*39*/ NULL,
1131: NULL,
1132: NULL,
1133: MatGetValues_MPISELL,
1134: MatCopy_MPISELL,
1135: /*44*/ NULL,
1136: MatScale_MPISELL,
1137: MatShift_MPISELL,
1138: MatDiagonalSet_MPISELL,
1139: NULL,
1140: /*49*/ MatSetRandom_MPISELL,
1141: NULL,
1142: NULL,
1143: NULL,
1144: NULL,
1145: /*54*/ MatFDColoringCreate_MPIXAIJ,
1146: NULL,
1147: MatSetUnfactored_MPISELL,
1148: NULL,
1149: NULL,
1150: /*59*/ NULL,
1151: MatDestroy_MPISELL,
1152: MatView_MPISELL,
1153: NULL,
1154: NULL,
1155: /*64*/ NULL,
1156: NULL,
1157: NULL,
1158: NULL,
1159: NULL,
1160: /*69*/ NULL,
1161: NULL,
1162: NULL,
1163: MatFDColoringApply_AIJ, /* reuse AIJ function */
1164: MatSetFromOptions_MPISELL,
1165: NULL,
1166: /*75*/ NULL,
1167: NULL,
1168: NULL,
1169: NULL,
1170: NULL,
1171: /*80*/ NULL,
1172: NULL,
1173: NULL,
1174: /*83*/ NULL,
1175: NULL,
1176: NULL,
1177: NULL,
1178: NULL,
1179: NULL,
1180: /*89*/ NULL,
1181: NULL,
1182: NULL,
1183: NULL,
1184: MatConjugate_MPISELL,
1185: /*94*/ NULL,
1186: NULL,
1187: MatRealPart_MPISELL,
1188: MatImaginaryPart_MPISELL,
1189: NULL,
1190: /*99*/ NULL,
1191: NULL,
1192: NULL,
1193: NULL,
1194: NULL,
1195: /*104*/ MatMissingDiagonal_MPISELL,
1196: NULL,
1197: NULL,
1198: MatGetGhosts_MPISELL,
1199: NULL,
1200: /*109*/ NULL,
1201: MatMultDiagonalBlock_MPISELL,
1202: NULL,
1203: NULL,
1204: NULL,
1205: /*114*/ NULL,
1206: NULL,
1207: NULL,
1208: MatInvertBlockDiagonal_MPISELL,
1209: NULL,
1210: /*119*/ NULL,
1211: NULL,
1212: NULL,
1213: NULL,
1214: NULL,
1215: /*124*/ NULL,
1216: NULL,
1217: NULL,
1218: NULL,
1219: NULL,
1220: /*129*/ MatFDColoringSetUp_MPIXAIJ,
1221: NULL,
1222: NULL,
1223: NULL,
1224: NULL,
1225: /*134*/ NULL,
1226: NULL,
1227: NULL,
1228: NULL,
1229: NULL,
1230: /*139*/ NULL,
1231: NULL,
1232: NULL,
1233: NULL};
1235: /*@C
1236: MatMPISELLSetPreallocation - Preallocates memory for a `MATMPISELL` sparse parallel matrix in sell format.
1237: For good matrix assembly performance the user should preallocate the matrix storage by
1238: setting the parameters `d_nz` (or `d_nnz`) and `o_nz` (or `o_nnz`).
1240: Collective
1242: Input Parameters:
1243: + B - the matrix
1244: . d_nz - number of nonzeros per row in DIAGONAL portion of local submatrix
1245: (same value is used for all local rows)
1246: . d_nnz - array containing the number of nonzeros in the various rows of the
1247: DIAGONAL portion of the local submatrix (possibly different for each row)
1248: or NULL (`PETSC_NULL_INTEGER` in Fortran), if `d_nz` is used to specify the nonzero structure.
1249: The size of this array is equal to the number of local rows, i.e 'm'.
1250: For matrices that will be factored, you must leave room for (and set)
1251: the diagonal entry even if it is zero.
1252: . o_nz - number of nonzeros per row in the OFF-DIAGONAL portion of local
1253: submatrix (same value is used for all local rows).
1254: - o_nnz - array containing the number of nonzeros in the various rows of the
1255: OFF-DIAGONAL portion of the local submatrix (possibly different for
1256: each row) or NULL (`PETSC_NULL_INTEGER` in Fortran), if `o_nz` is used to specify the nonzero
1257: structure. The size of this array is equal to the number
1258: of local rows, i.e 'm'.
1260: Example usage:
1261: Consider the following 8x8 matrix with 34 non-zero values, that is
1262: assembled across 3 processors. Lets assume that proc0 owns 3 rows,
1263: proc1 owns 3 rows, proc2 owns 2 rows. This division can be shown
1264: as follows
1266: .vb
1267: 1 2 0 | 0 3 0 | 0 4
1268: Proc0 0 5 6 | 7 0 0 | 8 0
1269: 9 0 10 | 11 0 0 | 12 0
1270: -------------------------------------
1271: 13 0 14 | 15 16 17 | 0 0
1272: Proc1 0 18 0 | 19 20 21 | 0 0
1273: 0 0 0 | 22 23 0 | 24 0
1274: -------------------------------------
1275: Proc2 25 26 27 | 0 0 28 | 29 0
1276: 30 0 0 | 31 32 33 | 0 34
1277: .ve
1279: This can be represented as a collection of submatrices as
1281: .vb
1282: A B C
1283: D E F
1284: G H I
1285: .ve
1287: Where the submatrices A,B,C are owned by proc0, D,E,F are
1288: owned by proc1, G,H,I are owned by proc2.
1290: The 'm' parameters for proc0,proc1,proc2 are 3,3,2 respectively.
1291: The 'n' parameters for proc0,proc1,proc2 are 3,3,2 respectively.
1292: The 'M','N' parameters are 8,8, and have the same values on all procs.
1294: The DIAGONAL submatrices corresponding to proc0,proc1,proc2 are
1295: submatrices [A], [E], [I] respectively. The OFF-DIAGONAL submatrices
1296: corresponding to proc0,proc1,proc2 are [BC], [DF], [GH] respectively.
1297: Internally, each processor stores the DIAGONAL part, and the OFF-DIAGONAL
1298: part as `MATSEQSELL` matrices. For example, proc1 will store [E] as a `MATSEQSELL`
1299: matrix, and [DF] as another SeqSELL matrix.
1301: When `d_nz`, `o_nz` parameters are specified, `d_nz` storage elements are
1302: allocated for every row of the local DIAGONAL submatrix, and o_nz
1303: storage locations are allocated for every row of the OFF-DIAGONAL submatrix.
1304: One way to choose `d_nz` and `o_nz` is to use the maximum number of nonzeros over
1305: the local rows for each of the local DIAGONAL, and the OFF-DIAGONAL submatrices.
1306: In this case, the values of d_nz,o_nz are
1307: .vb
1308: proc0 dnz = 2, o_nz = 2
1309: proc1 dnz = 3, o_nz = 2
1310: proc2 dnz = 1, o_nz = 4
1311: .ve
1312: We are allocating m*(d_nz+o_nz) storage locations for every proc. This
1313: translates to 3*(2+2)=12 for proc0, 3*(3+2)=15 for proc1, 2*(1+4)=10
1314: for proc3. i.e we are using 12+15+10=37 storage locations to store
1315: 34 values.
1317: When `d_nnz`, `o_nnz` parameters are specified, the storage is specified
1318: for every row, corresponding to both DIAGONAL and OFF-DIAGONAL submatrices.
1319: In the above case the values for d_nnz,o_nnz are
1320: .vb
1321: proc0 d_nnz = [2,2,2] and o_nnz = [2,2,2]
1322: proc1 d_nnz = [3,3,2] and o_nnz = [2,1,1]
1323: proc2 d_nnz = [1,1] and o_nnz = [4,4]
1324: .ve
1325: Here the space allocated is according to nz (or maximum values in the nnz
1326: if nnz is provided) for DIAGONAL and OFF-DIAGONAL submatrices, i.e (2+2+3+2)*3+(1+4)*2=37
1328: Level: intermediate
1330: Notes:
1331: If the *_nnz parameter is given then the *_nz parameter is ignored
1333: The stored row and column indices begin with zero.
1335: The parallel matrix is partitioned such that the first m0 rows belong to
1336: process 0, the next m1 rows belong to process 1, the next m2 rows belong
1337: to process 2 etc.. where m0,m1,m2... are the input parameter 'm'.
1339: The DIAGONAL portion of the local submatrix of a processor can be defined
1340: as the submatrix which is obtained by extraction the part corresponding to
1341: the rows r1-r2 and columns c1-c2 of the global matrix, where r1 is the
1342: first row that belongs to the processor, r2 is the last row belonging to
1343: the this processor, and c1-c2 is range of indices of the local part of a
1344: vector suitable for applying the matrix to. This is an mxn matrix. In the
1345: common case of a square matrix, the row and column ranges are the same and
1346: the DIAGONAL part is also square. The remaining portion of the local
1347: submatrix (mxN) constitute the OFF-DIAGONAL portion.
1349: If `o_nnz`, `d_nnz` are specified, then `o_nz`, and `d_nz` are ignored.
1351: You can call `MatGetInfo()` to get information on how effective the preallocation was;
1352: for example the fields mallocs,nz_allocated,nz_used,nz_unneeded;
1353: You can also run with the option -info and look for messages with the string
1354: malloc in them to see if additional memory allocation was needed.
1356: .seealso: `Mat`, `MatCreate()`, `MatCreateSeqSELL()`, `MatSetValues()`, `MatCreateSELL()`,
1357: `MATMPISELL`, `MatGetInfo()`, `PetscSplitOwnership()`, `MATSELL`
1358: @*/
1359: PetscErrorCode MatMPISELLSetPreallocation(Mat B, PetscInt d_nz, const PetscInt d_nnz[], PetscInt o_nz, const PetscInt o_nnz[])
1360: {
1361: PetscFunctionBegin;
1364: PetscTryMethod(B, "MatMPISELLSetPreallocation_C", (Mat, PetscInt, const PetscInt[], PetscInt, const PetscInt[]), (B, d_nz, d_nnz, o_nz, o_nnz));
1365: PetscFunctionReturn(PETSC_SUCCESS);
1366: }
1368: /*MC
1369: MATMPISELL - MATMPISELL = "mpisell" - A matrix type to be used for MPI sparse matrices,
1370: based on the sliced Ellpack format
1372: Options Database Key:
1373: . -mat_type sell - sets the matrix type to `MATSELL` during a call to `MatSetFromOptions()`
1375: Level: beginner
1377: .seealso: `Mat`, `MatCreateSELL()`, `MATSEQSELL`, `MATSELL`, `MATSEQAIJ`, `MATAIJ`, `MATMPIAIJ`
1378: M*/
1380: /*@C
1381: MatCreateSELL - Creates a sparse parallel matrix in `MATSELL` format.
1383: Collective
1385: Input Parameters:
1386: + comm - MPI communicator
1387: . m - number of local rows (or `PETSC_DECIDE` to have calculated if M is given)
1388: This value should be the same as the local size used in creating the
1389: y vector for the matrix-vector product y = Ax.
1390: . n - This value should be the same as the local size used in creating the
1391: x vector for the matrix-vector product y = Ax. (or `PETSC_DECIDE` to have
1392: calculated if `N` is given) For square matrices n is almost always `m`.
1393: . M - number of global rows (or `PETSC_DETERMINE` to have calculated if `m` is given)
1394: . N - number of global columns (or `PETSC_DETERMINE` to have calculated if `n` is given)
1395: . d_rlenmax - max number of nonzeros per row in DIAGONAL portion of local submatrix
1396: (same value is used for all local rows)
1397: . d_rlen - array containing the number of nonzeros in the various rows of the
1398: DIAGONAL portion of the local submatrix (possibly different for each row)
1399: or `NULL`, if d_rlenmax is used to specify the nonzero structure.
1400: The size of this array is equal to the number of local rows, i.e `m`.
1401: . o_rlenmax - max number of nonzeros per row in the OFF-DIAGONAL portion of local
1402: submatrix (same value is used for all local rows).
1403: - o_rlen - array containing the number of nonzeros in the various rows of the
1404: OFF-DIAGONAL portion of the local submatrix (possibly different for
1405: each row) or `NULL`, if `o_rlenmax` is used to specify the nonzero
1406: structure. The size of this array is equal to the number
1407: of local rows, i.e `m`.
1409: Output Parameter:
1410: . A - the matrix
1412: Options Database Key:
1413: . -mat_sell_oneindex - Internally use indexing starting at 1
1414: rather than 0. When calling `MatSetValues()`,
1415: the user still MUST index entries starting at 0!
1417: Example:
1418: Consider the following 8x8 matrix with 34 non-zero values, that is
1419: assembled across 3 processors. Lets assume that proc0 owns 3 rows,
1420: proc1 owns 3 rows, proc2 owns 2 rows. This division can be shown
1421: as follows
1423: .vb
1424: 1 2 0 | 0 3 0 | 0 4
1425: Proc0 0 5 6 | 7 0 0 | 8 0
1426: 9 0 10 | 11 0 0 | 12 0
1427: -------------------------------------
1428: 13 0 14 | 15 16 17 | 0 0
1429: Proc1 0 18 0 | 19 20 21 | 0 0
1430: 0 0 0 | 22 23 0 | 24 0
1431: -------------------------------------
1432: Proc2 25 26 27 | 0 0 28 | 29 0
1433: 30 0 0 | 31 32 33 | 0 34
1434: .ve
1436: This can be represented as a collection of submatrices as
1437: .vb
1438: A B C
1439: D E F
1440: G H I
1441: .ve
1443: Where the submatrices A,B,C are owned by proc0, D,E,F are
1444: owned by proc1, G,H,I are owned by proc2.
1446: The 'm' parameters for proc0,proc1,proc2 are 3,3,2 respectively.
1447: The 'n' parameters for proc0,proc1,proc2 are 3,3,2 respectively.
1448: The 'M','N' parameters are 8,8, and have the same values on all procs.
1450: The DIAGONAL submatrices corresponding to proc0,proc1,proc2 are
1451: submatrices [A], [E], [I] respectively. The OFF-DIAGONAL submatrices
1452: corresponding to proc0,proc1,proc2 are [BC], [DF], [GH] respectively.
1453: Internally, each processor stores the DIAGONAL part, and the OFF-DIAGONAL
1454: part as `MATSEQSELL` matrices. For example, proc1 will store [E] as a `MATSEQSELL`
1455: matrix, and [DF] as another `MATSEQSELL` matrix.
1457: When d_rlenmax, o_rlenmax parameters are specified, d_rlenmax storage elements are
1458: allocated for every row of the local DIAGONAL submatrix, and o_rlenmax
1459: storage locations are allocated for every row of the OFF-DIAGONAL submatrix.
1460: One way to choose `d_rlenmax` and `o_rlenmax` is to use the maximum number of nonzeros over
1461: the local rows for each of the local DIAGONAL, and the OFF-DIAGONAL submatrices.
1462: In this case, the values of d_rlenmax,o_rlenmax are
1463: .vb
1464: proc0 - d_rlenmax = 2, o_rlenmax = 2
1465: proc1 - d_rlenmax = 3, o_rlenmax = 2
1466: proc2 - d_rlenmax = 1, o_rlenmax = 4
1467: .ve
1468: We are allocating m*(d_rlenmax+o_rlenmax) storage locations for every proc. This
1469: translates to 3*(2+2)=12 for proc0, 3*(3+2)=15 for proc1, 2*(1+4)=10
1470: for proc3. i.e we are using 12+15+10=37 storage locations to store
1471: 34 values.
1473: When `d_rlen`, `o_rlen` parameters are specified, the storage is specified
1474: for every row, corresponding to both DIAGONAL and OFF-DIAGONAL submatrices.
1475: In the above case the values for `d_nnz`, `o_nnz` are
1476: .vb
1477: proc0 - d_nnz = [2,2,2] and o_nnz = [2,2,2]
1478: proc1 - d_nnz = [3,3,2] and o_nnz = [2,1,1]
1479: proc2 - d_nnz = [1,1] and o_nnz = [4,4]
1480: .ve
1481: Here the space allocated is still 37 though there are 34 nonzeros because
1482: the allocation is always done according to rlenmax.
1484: Level: intermediate
1486: Notes:
1487: It is recommended that one use the `MatCreate()`, `MatSetType()` and/or `MatSetFromOptions()`,
1488: MatXXXXSetPreallocation() paradigm instead of this routine directly.
1489: [MatXXXXSetPreallocation() is, for example, `MatSeqSELLSetPreallocation()`]
1491: If the *_rlen parameter is given then the *_rlenmax parameter is ignored
1493: `m`, `n`, `M`, `N` parameters specify the size of the matrix, and its partitioning across
1494: processors, while `d_rlenmax`, `d_rlen`, `o_rlenmax` , `o_rlen` parameters specify the approximate
1495: storage requirements for this matrix.
1497: If `PETSC_DECIDE` or `PETSC_DETERMINE` is used for a particular argument on one
1498: processor than it must be used on all processors that share the object for
1499: that argument.
1501: The user MUST specify either the local or global matrix dimensions
1502: (possibly both).
1504: The parallel matrix is partitioned across processors such that the
1505: first m0 rows belong to process 0, the next m1 rows belong to
1506: process 1, the next m2 rows belong to process 2 etc.. where
1507: m0,m1,m2,.. are the input parameter 'm'. i.e each processor stores
1508: values corresponding to [`m` x `N`] submatrix.
1510: The columns are logically partitioned with the n0 columns belonging
1511: to 0th partition, the next n1 columns belonging to the next
1512: partition etc.. where n0,n1,n2... are the input parameter `n`.
1514: The DIAGONAL portion of the local submatrix on any given processor
1515: is the submatrix corresponding to the rows and columns `m`, `n`
1516: corresponding to the given processor. i.e diagonal matrix on
1517: process 0 is [m0 x n0], diagonal matrix on process 1 is [m1 x n1]
1518: etc. The remaining portion of the local submatrix [m x (N-n)]
1519: constitute the OFF-DIAGONAL portion. The example below better
1520: illustrates this concept.
1522: For a square global matrix we define each processor's diagonal portion
1523: to be its local rows and the corresponding columns (a square submatrix);
1524: each processor's off-diagonal portion encompasses the remainder of the
1525: local matrix (a rectangular submatrix).
1527: If `o_rlen`, `d_rlen` are specified, then `o_rlenmax`, and `d_rlenmax` are ignored.
1529: When calling this routine with a single process communicator, a matrix of
1530: type `MATSEQSELL` is returned. If a matrix of type `MATMPISELL` is desired for this
1531: type of communicator, use the construction mechanism
1532: .vb
1533: MatCreate(...,&A);
1534: MatSetType(A,MATMPISELL);
1535: MatSetSizes(A, m,n,M,N);
1536: MatMPISELLSetPreallocation(A,...);
1537: .ve
1539: .seealso: `Mat`, `MATSELL`, `MatCreate()`, `MatCreateSeqSELL()`, `MatSetValues()`, `MatMPISELLSetPreallocation()`, `MATMPISELL`
1540: @*/
1541: PetscErrorCode MatCreateSELL(MPI_Comm comm, PetscInt m, PetscInt n, PetscInt M, PetscInt N, PetscInt d_rlenmax, const PetscInt d_rlen[], PetscInt o_rlenmax, const PetscInt o_rlen[], Mat *A)
1542: {
1543: PetscMPIInt size;
1545: PetscFunctionBegin;
1546: PetscCall(MatCreate(comm, A));
1547: PetscCall(MatSetSizes(*A, m, n, M, N));
1548: PetscCallMPI(MPI_Comm_size(comm, &size));
1549: if (size > 1) {
1550: PetscCall(MatSetType(*A, MATMPISELL));
1551: PetscCall(MatMPISELLSetPreallocation(*A, d_rlenmax, d_rlen, o_rlenmax, o_rlen));
1552: } else {
1553: PetscCall(MatSetType(*A, MATSEQSELL));
1554: PetscCall(MatSeqSELLSetPreallocation(*A, d_rlenmax, d_rlen));
1555: }
1556: PetscFunctionReturn(PETSC_SUCCESS);
1557: }
1559: /*@C
1560: MatMPISELLGetSeqSELL - Returns the local pieces of this distributed matrix
1562: Not Collective
1564: Input Parameter:
1565: . A - the `MATMPISELL` matrix
1567: Output Parameters:
1568: + Ad - The diagonal portion of `A`
1569: . Ao - The off-diagonal portion of `A`
1570: - colmap - An array mapping local column numbers of `Ao` to global column numbers of the parallel matrix
1572: Level: advanced
1574: .seealso: `Mat`, `MATSEQSELL`, `MATMPISELL`
1575: @*/
1576: PetscErrorCode MatMPISELLGetSeqSELL(Mat A, Mat *Ad, Mat *Ao, const PetscInt *colmap[])
1577: {
1578: Mat_MPISELL *a = (Mat_MPISELL *)A->data;
1579: PetscBool flg;
1581: PetscFunctionBegin;
1582: PetscCall(PetscObjectTypeCompare((PetscObject)A, MATMPISELL, &flg));
1583: PetscCheck(flg, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "This function requires a MATMPISELL matrix as input");
1584: if (Ad) *Ad = a->A;
1585: if (Ao) *Ao = a->B;
1586: if (colmap) *colmap = a->garray;
1587: PetscFunctionReturn(PETSC_SUCCESS);
1588: }
1590: /*@C
1591: MatMPISELLGetLocalMatCondensed - Creates a `MATSEQSELL` matrix from an `MATMPISELL` matrix by
1592: taking all its local rows and NON-ZERO columns
1594: Not Collective
1596: Input Parameters:
1597: + A - the matrix
1598: . scall - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
1599: . row - index sets of rows to extract (or `NULL`)
1600: - col - index sets of columns to extract (or `NULL`)
1602: Output Parameter:
1603: . A_loc - the local sequential matrix generated
1605: Level: advanced
1607: .seealso: `Mat`, `MATSEQSELL`, `MATMPISELL`, `MatGetOwnershipRange()`, `MatMPISELLGetLocalMat()`
1608: @*/
1609: PetscErrorCode MatMPISELLGetLocalMatCondensed(Mat A, MatReuse scall, IS *row, IS *col, Mat *A_loc)
1610: {
1611: Mat_MPISELL *a = (Mat_MPISELL *)A->data;
1612: PetscInt i, start, end, ncols, nzA, nzB, *cmap, imark, *idx;
1613: IS isrowa, iscola;
1614: Mat *aloc;
1615: PetscBool match;
1617: PetscFunctionBegin;
1618: PetscCall(PetscObjectTypeCompare((PetscObject)A, MATMPISELL, &match));
1619: PetscCheck(match, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Requires MATMPISELL matrix as input");
1620: PetscCall(PetscLogEventBegin(MAT_Getlocalmatcondensed, A, 0, 0, 0));
1621: if (!row) {
1622: start = A->rmap->rstart;
1623: end = A->rmap->rend;
1624: PetscCall(ISCreateStride(PETSC_COMM_SELF, end - start, start, 1, &isrowa));
1625: } else {
1626: isrowa = *row;
1627: }
1628: if (!col) {
1629: start = A->cmap->rstart;
1630: cmap = a->garray;
1631: nzA = a->A->cmap->n;
1632: nzB = a->B->cmap->n;
1633: PetscCall(PetscMalloc1(nzA + nzB, &idx));
1634: ncols = 0;
1635: for (i = 0; i < nzB; i++) {
1636: if (cmap[i] < start) idx[ncols++] = cmap[i];
1637: else break;
1638: }
1639: imark = i;
1640: for (i = 0; i < nzA; i++) idx[ncols++] = start + i;
1641: for (i = imark; i < nzB; i++) idx[ncols++] = cmap[i];
1642: PetscCall(ISCreateGeneral(PETSC_COMM_SELF, ncols, idx, PETSC_OWN_POINTER, &iscola));
1643: } else {
1644: iscola = *col;
1645: }
1646: if (scall != MAT_INITIAL_MATRIX) {
1647: PetscCall(PetscMalloc1(1, &aloc));
1648: aloc[0] = *A_loc;
1649: }
1650: PetscCall(MatCreateSubMatrices(A, 1, &isrowa, &iscola, scall, &aloc));
1651: *A_loc = aloc[0];
1652: PetscCall(PetscFree(aloc));
1653: if (!row) PetscCall(ISDestroy(&isrowa));
1654: if (!col) PetscCall(ISDestroy(&iscola));
1655: PetscCall(PetscLogEventEnd(MAT_Getlocalmatcondensed, A, 0, 0, 0));
1656: PetscFunctionReturn(PETSC_SUCCESS);
1657: }
1659: #include <../src/mat/impls/aij/mpi/mpiaij.h>
1661: PetscErrorCode MatConvert_MPISELL_MPIAIJ(Mat A, MatType newtype, MatReuse reuse, Mat *newmat)
1662: {
1663: Mat_MPISELL *a = (Mat_MPISELL *)A->data;
1664: Mat B;
1665: Mat_MPIAIJ *b;
1667: PetscFunctionBegin;
1668: PetscCheck(A->assembled, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Matrix must be assembled");
1670: if (reuse == MAT_REUSE_MATRIX) {
1671: B = *newmat;
1672: } else {
1673: PetscCall(MatCreate(PetscObjectComm((PetscObject)A), &B));
1674: PetscCall(MatSetType(B, MATMPIAIJ));
1675: PetscCall(MatSetSizes(B, A->rmap->n, A->cmap->n, A->rmap->N, A->cmap->N));
1676: PetscCall(MatSetBlockSizes(B, A->rmap->bs, A->cmap->bs));
1677: PetscCall(MatSeqAIJSetPreallocation(B, 0, NULL));
1678: PetscCall(MatMPIAIJSetPreallocation(B, 0, NULL, 0, NULL));
1679: }
1680: b = (Mat_MPIAIJ *)B->data;
1682: if (reuse == MAT_REUSE_MATRIX) {
1683: PetscCall(MatConvert_SeqSELL_SeqAIJ(a->A, MATSEQAIJ, MAT_REUSE_MATRIX, &b->A));
1684: PetscCall(MatConvert_SeqSELL_SeqAIJ(a->B, MATSEQAIJ, MAT_REUSE_MATRIX, &b->B));
1685: } else {
1686: PetscCall(MatDestroy(&b->A));
1687: PetscCall(MatDestroy(&b->B));
1688: PetscCall(MatDisAssemble_MPISELL(A));
1689: PetscCall(MatConvert_SeqSELL_SeqAIJ(a->A, MATSEQAIJ, MAT_INITIAL_MATRIX, &b->A));
1690: PetscCall(MatConvert_SeqSELL_SeqAIJ(a->B, MATSEQAIJ, MAT_INITIAL_MATRIX, &b->B));
1691: PetscCall(MatAssemblyBegin(B, MAT_FINAL_ASSEMBLY));
1692: PetscCall(MatAssemblyEnd(B, MAT_FINAL_ASSEMBLY));
1693: PetscCall(MatAssemblyBegin(A, MAT_FINAL_ASSEMBLY));
1694: PetscCall(MatAssemblyEnd(A, MAT_FINAL_ASSEMBLY));
1695: }
1697: if (reuse == MAT_INPLACE_MATRIX) {
1698: PetscCall(MatHeaderReplace(A, &B));
1699: } else {
1700: *newmat = B;
1701: }
1702: PetscFunctionReturn(PETSC_SUCCESS);
1703: }
1705: PetscErrorCode MatConvert_MPIAIJ_MPISELL(Mat A, MatType newtype, MatReuse reuse, Mat *newmat)
1706: {
1707: Mat_MPIAIJ *a = (Mat_MPIAIJ *)A->data;
1708: Mat B;
1709: Mat_MPISELL *b;
1711: PetscFunctionBegin;
1712: PetscCheck(A->assembled, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Matrix must be assembled");
1714: if (reuse == MAT_REUSE_MATRIX) {
1715: B = *newmat;
1716: } else {
1717: Mat_SeqAIJ *Aa = (Mat_SeqAIJ *)a->A->data, *Ba = (Mat_SeqAIJ *)a->B->data;
1718: PetscInt i, d_nz = 0, o_nz = 0, m = A->rmap->N, n = A->cmap->N, lm = A->rmap->n, ln = A->cmap->n;
1719: PetscInt *d_nnz, *o_nnz;
1720: PetscCall(PetscMalloc2(lm, &d_nnz, lm, &o_nnz));
1721: for (i = 0; i < lm; i++) {
1722: d_nnz[i] = Aa->i[i + 1] - Aa->i[i];
1723: o_nnz[i] = Ba->i[i + 1] - Ba->i[i];
1724: if (d_nnz[i] > d_nz) d_nz = d_nnz[i];
1725: if (o_nnz[i] > o_nz) o_nz = o_nnz[i];
1726: }
1727: PetscCall(MatCreate(PetscObjectComm((PetscObject)A), &B));
1728: PetscCall(MatSetType(B, MATMPISELL));
1729: PetscCall(MatSetSizes(B, lm, ln, m, n));
1730: PetscCall(MatSetBlockSizes(B, A->rmap->bs, A->cmap->bs));
1731: PetscCall(MatSeqSELLSetPreallocation(B, d_nz, d_nnz));
1732: PetscCall(MatMPISELLSetPreallocation(B, d_nz, d_nnz, o_nz, o_nnz));
1733: PetscCall(PetscFree2(d_nnz, o_nnz));
1734: }
1735: b = (Mat_MPISELL *)B->data;
1737: if (reuse == MAT_REUSE_MATRIX) {
1738: PetscCall(MatConvert_SeqAIJ_SeqSELL(a->A, MATSEQSELL, MAT_REUSE_MATRIX, &b->A));
1739: PetscCall(MatConvert_SeqAIJ_SeqSELL(a->B, MATSEQSELL, MAT_REUSE_MATRIX, &b->B));
1740: } else {
1741: PetscCall(MatDestroy(&b->A));
1742: PetscCall(MatDestroy(&b->B));
1743: PetscCall(MatConvert_SeqAIJ_SeqSELL(a->A, MATSEQSELL, MAT_INITIAL_MATRIX, &b->A));
1744: PetscCall(MatConvert_SeqAIJ_SeqSELL(a->B, MATSEQSELL, MAT_INITIAL_MATRIX, &b->B));
1745: PetscCall(MatAssemblyBegin(A, MAT_FINAL_ASSEMBLY));
1746: PetscCall(MatAssemblyEnd(A, MAT_FINAL_ASSEMBLY));
1747: PetscCall(MatAssemblyBegin(B, MAT_FINAL_ASSEMBLY));
1748: PetscCall(MatAssemblyEnd(B, MAT_FINAL_ASSEMBLY));
1749: }
1751: if (reuse == MAT_INPLACE_MATRIX) {
1752: PetscCall(MatHeaderReplace(A, &B));
1753: } else {
1754: *newmat = B;
1755: }
1756: PetscFunctionReturn(PETSC_SUCCESS);
1757: }
1759: PetscErrorCode MatSOR_MPISELL(Mat matin, Vec bb, PetscReal omega, MatSORType flag, PetscReal fshift, PetscInt its, PetscInt lits, Vec xx)
1760: {
1761: Mat_MPISELL *mat = (Mat_MPISELL *)matin->data;
1762: Vec bb1 = NULL;
1764: PetscFunctionBegin;
1765: if (flag == SOR_APPLY_UPPER) {
1766: PetscCall((*mat->A->ops->sor)(mat->A, bb, omega, flag, fshift, lits, 1, xx));
1767: PetscFunctionReturn(PETSC_SUCCESS);
1768: }
1770: if (its > 1 || ~flag & SOR_ZERO_INITIAL_GUESS || flag & SOR_EISENSTAT) PetscCall(VecDuplicate(bb, &bb1));
1772: if ((flag & SOR_LOCAL_SYMMETRIC_SWEEP) == SOR_LOCAL_SYMMETRIC_SWEEP) {
1773: if (flag & SOR_ZERO_INITIAL_GUESS) {
1774: PetscCall((*mat->A->ops->sor)(mat->A, bb, omega, flag, fshift, lits, 1, xx));
1775: its--;
1776: }
1778: while (its--) {
1779: PetscCall(VecScatterBegin(mat->Mvctx, xx, mat->lvec, INSERT_VALUES, SCATTER_FORWARD));
1780: PetscCall(VecScatterEnd(mat->Mvctx, xx, mat->lvec, INSERT_VALUES, SCATTER_FORWARD));
1782: /* update rhs: bb1 = bb - B*x */
1783: PetscCall(VecScale(mat->lvec, -1.0));
1784: PetscCall((*mat->B->ops->multadd)(mat->B, mat->lvec, bb, bb1));
1786: /* local sweep */
1787: PetscCall((*mat->A->ops->sor)(mat->A, bb1, omega, SOR_SYMMETRIC_SWEEP, fshift, lits, 1, xx));
1788: }
1789: } else if (flag & SOR_LOCAL_FORWARD_SWEEP) {
1790: if (flag & SOR_ZERO_INITIAL_GUESS) {
1791: PetscCall((*mat->A->ops->sor)(mat->A, bb, omega, flag, fshift, lits, 1, xx));
1792: its--;
1793: }
1794: while (its--) {
1795: PetscCall(VecScatterBegin(mat->Mvctx, xx, mat->lvec, INSERT_VALUES, SCATTER_FORWARD));
1796: PetscCall(VecScatterEnd(mat->Mvctx, xx, mat->lvec, INSERT_VALUES, SCATTER_FORWARD));
1798: /* update rhs: bb1 = bb - B*x */
1799: PetscCall(VecScale(mat->lvec, -1.0));
1800: PetscCall((*mat->B->ops->multadd)(mat->B, mat->lvec, bb, bb1));
1802: /* local sweep */
1803: PetscCall((*mat->A->ops->sor)(mat->A, bb1, omega, SOR_FORWARD_SWEEP, fshift, lits, 1, xx));
1804: }
1805: } else if (flag & SOR_LOCAL_BACKWARD_SWEEP) {
1806: if (flag & SOR_ZERO_INITIAL_GUESS) {
1807: PetscCall((*mat->A->ops->sor)(mat->A, bb, omega, flag, fshift, lits, 1, xx));
1808: its--;
1809: }
1810: while (its--) {
1811: PetscCall(VecScatterBegin(mat->Mvctx, xx, mat->lvec, INSERT_VALUES, SCATTER_FORWARD));
1812: PetscCall(VecScatterEnd(mat->Mvctx, xx, mat->lvec, INSERT_VALUES, SCATTER_FORWARD));
1814: /* update rhs: bb1 = bb - B*x */
1815: PetscCall(VecScale(mat->lvec, -1.0));
1816: PetscCall((*mat->B->ops->multadd)(mat->B, mat->lvec, bb, bb1));
1818: /* local sweep */
1819: PetscCall((*mat->A->ops->sor)(mat->A, bb1, omega, SOR_BACKWARD_SWEEP, fshift, lits, 1, xx));
1820: }
1821: } else SETERRQ(PetscObjectComm((PetscObject)matin), PETSC_ERR_SUP, "Parallel SOR not supported");
1823: PetscCall(VecDestroy(&bb1));
1825: matin->factorerrortype = mat->A->factorerrortype;
1826: PetscFunctionReturn(PETSC_SUCCESS);
1827: }
1829: #if defined(PETSC_HAVE_CUDA)
1830: PETSC_INTERN PetscErrorCode MatConvert_MPISELL_MPISELLCUDA(Mat, MatType, MatReuse, Mat *);
1831: #endif
1833: /*MC
1834: MATMPISELL - MATMPISELL = "MPISELL" - A matrix type to be used for parallel sparse matrices.
1836: Options Database Keys:
1837: . -mat_type mpisell - sets the matrix type to `MATMPISELL` during a call to `MatSetFromOptions()`
1839: Level: beginner
1841: .seealso: `Mat`, `MATSELL`, `MATSEQSELL` `MatCreateSELL()`
1842: M*/
1843: PETSC_EXTERN PetscErrorCode MatCreate_MPISELL(Mat B)
1844: {
1845: Mat_MPISELL *b;
1846: PetscMPIInt size;
1848: PetscFunctionBegin;
1849: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)B), &size));
1850: PetscCall(PetscNew(&b));
1851: B->data = (void *)b;
1852: B->ops[0] = MatOps_Values;
1853: B->assembled = PETSC_FALSE;
1854: B->insertmode = NOT_SET_VALUES;
1855: b->size = size;
1856: PetscCallMPI(MPI_Comm_rank(PetscObjectComm((PetscObject)B), &b->rank));
1857: /* build cache for off array entries formed */
1858: PetscCall(MatStashCreate_Private(PetscObjectComm((PetscObject)B), 1, &B->stash));
1860: b->donotstash = PETSC_FALSE;
1861: b->colmap = NULL;
1862: b->garray = NULL;
1863: b->roworiented = PETSC_TRUE;
1865: /* stuff used for matrix vector multiply */
1866: b->lvec = NULL;
1867: b->Mvctx = NULL;
1869: /* stuff for MatGetRow() */
1870: b->rowindices = NULL;
1871: b->rowvalues = NULL;
1872: b->getrowactive = PETSC_FALSE;
1874: PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatStoreValues_C", MatStoreValues_MPISELL));
1875: PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatRetrieveValues_C", MatRetrieveValues_MPISELL));
1876: PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatIsTranspose_C", MatIsTranspose_MPISELL));
1877: PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatMPISELLSetPreallocation_C", MatMPISELLSetPreallocation_MPISELL));
1878: PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatConvert_mpisell_mpiaij_C", MatConvert_MPISELL_MPIAIJ));
1879: #if defined(PETSC_HAVE_CUDA)
1880: PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatConvert_mpisell_mpisellcuda_C", MatConvert_MPISELL_MPISELLCUDA));
1881: #endif
1882: PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatDiagonalScaleLocal_C", MatDiagonalScaleLocal_MPISELL));
1883: PetscCall(PetscObjectChangeTypeName((PetscObject)B, MATMPISELL));
1884: PetscFunctionReturn(PETSC_SUCCESS);
1885: }