Actual source code: pipelcg.c
1: #include <petsc/private/kspimpl.h>
2: #include <petsc/private/vecimpl.h>
4: #define offset(j) PetscMax(((j) - (2 * l)), 0)
5: #define shift(i, j) ((i) - offset(j))
6: #define G(i, j) (plcg->G[((j) * (2 * l + 1)) + (shift((i), (j)))])
7: #define G_noshift(i, j) (plcg->G[((j) * (2 * l + 1)) + (i)])
8: #define alpha(i) (plcg->alpha[i])
9: #define gamma(i) (plcg->gamma[i])
10: #define delta(i) (plcg->delta[i])
11: #define sigma(i) (plcg->sigma[i])
12: #define req(i) (plcg->req[i])
14: typedef struct KSP_CG_PIPE_L_s KSP_CG_PIPE_L;
15: struct KSP_CG_PIPE_L_s {
16: PetscInt l; /* pipeline depth */
17: Vec *Z; /* Z vectors (shifted base) */
18: Vec *U; /* U vectors (unpreconditioned shifted base) */
19: Vec *V; /* V vectors (original base) */
20: Vec *Q; /* Q vectors (auxiliary bases) */
21: Vec p; /* work vector */
22: PetscScalar *G; /* such that Z = VG (band matrix)*/
23: PetscScalar *gamma, *delta, *alpha;
24: PetscReal lmin, lmax; /* min and max eigen values estimates to compute base shifts */
25: PetscReal *sigma; /* base shifts */
26: MPI_Request *req; /* request array for asynchronous global collective */
27: PetscBool show_rstrt; /* flag to show restart information in output (default: not shown) */
28: };
30: /*
31: KSPSetUp_PIPELCG - Sets up the workspace needed by the PIPELCG method.
33: This is called once, usually automatically by KSPSolve() or KSPSetUp()
34: but can be called directly by KSPSetUp()
35: */
36: static PetscErrorCode KSPSetUp_PIPELCG(KSP ksp)
37: {
38: KSP_CG_PIPE_L *plcg = (KSP_CG_PIPE_L *)ksp->data;
39: PetscInt l = plcg->l, max_it = ksp->max_it;
40: MPI_Comm comm;
42: PetscFunctionBegin;
43: comm = PetscObjectComm((PetscObject)ksp);
44: PetscCheck(max_it >= 1, comm, PETSC_ERR_ARG_OUTOFRANGE, "%s: max_it argument must be positive.", ((PetscObject)ksp)->type_name);
45: PetscCheck(l >= 1, comm, PETSC_ERR_ARG_OUTOFRANGE, "%s: pipel argument must be positive.", ((PetscObject)ksp)->type_name);
46: PetscCheck(l <= max_it, comm, PETSC_ERR_ARG_OUTOFRANGE, "%s: pipel argument must be less than max_it.", ((PetscObject)ksp)->type_name);
48: PetscCall(KSPSetWorkVecs(ksp, 1)); /* get work vectors needed by PIPELCG */
49: plcg->p = ksp->work[0];
51: PetscCall(VecDuplicateVecs(plcg->p, PetscMax(3, l + 1), &plcg->Z));
52: PetscCall(VecDuplicateVecs(plcg->p, 3, &plcg->U));
53: PetscCall(VecDuplicateVecs(plcg->p, 3, &plcg->V));
54: PetscCall(VecDuplicateVecs(plcg->p, 3 * (l - 1) + 1, &plcg->Q));
55: PetscCall(PetscCalloc1(2, &plcg->alpha));
56: PetscCall(PetscCalloc1(l, &plcg->sigma));
57: PetscFunctionReturn(PETSC_SUCCESS);
58: }
60: static PetscErrorCode KSPReset_PIPELCG(KSP ksp)
61: {
62: KSP_CG_PIPE_L *plcg = (KSP_CG_PIPE_L *)ksp->data;
63: PetscInt l = plcg->l;
65: PetscFunctionBegin;
66: PetscCall(PetscFree(plcg->sigma));
67: PetscCall(PetscFree(plcg->alpha));
68: PetscCall(VecDestroyVecs(PetscMax(3, l + 1), &plcg->Z));
69: PetscCall(VecDestroyVecs(3, &plcg->U));
70: PetscCall(VecDestroyVecs(3, &plcg->V));
71: PetscCall(VecDestroyVecs(3 * (l - 1) + 1, &plcg->Q));
72: PetscFunctionReturn(PETSC_SUCCESS);
73: }
75: static PetscErrorCode KSPDestroy_PIPELCG(KSP ksp)
76: {
77: PetscFunctionBegin;
78: PetscCall(KSPReset_PIPELCG(ksp));
79: PetscCall(KSPDestroyDefault(ksp));
80: PetscFunctionReturn(PETSC_SUCCESS);
81: }
83: static PetscErrorCode KSPSetFromOptions_PIPELCG(KSP ksp, PetscOptionItems *PetscOptionsObject)
84: {
85: KSP_CG_PIPE_L *plcg = (KSP_CG_PIPE_L *)ksp->data;
86: PetscBool flag = PETSC_FALSE;
88: PetscFunctionBegin;
89: PetscOptionsHeadBegin(PetscOptionsObject, "KSP PIPELCG options");
90: PetscCall(PetscOptionsInt("-ksp_pipelcg_pipel", "Pipeline length", "", plcg->l, &plcg->l, &flag));
91: if (!flag) plcg->l = 1;
92: PetscCall(PetscOptionsReal("-ksp_pipelcg_lmin", "Estimate for smallest eigenvalue", "", plcg->lmin, &plcg->lmin, &flag));
93: if (!flag) plcg->lmin = 0.0;
94: PetscCall(PetscOptionsReal("-ksp_pipelcg_lmax", "Estimate for largest eigenvalue", "", plcg->lmax, &plcg->lmax, &flag));
95: if (!flag) plcg->lmax = 0.0;
96: PetscCall(PetscOptionsBool("-ksp_pipelcg_monitor", "Output information on restarts when they occur? (default: 0)", "", plcg->show_rstrt, &plcg->show_rstrt, &flag));
97: if (!flag) plcg->show_rstrt = PETSC_FALSE;
98: PetscOptionsHeadEnd();
99: PetscFunctionReturn(PETSC_SUCCESS);
100: }
102: static PetscErrorCode MPIPetsc_Iallreduce(void *sendbuf, void *recvbuf, PetscMPIInt count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request *request)
103: {
104: PetscFunctionBegin;
105: #if defined(PETSC_HAVE_MPI_NONBLOCKING_COLLECTIVES)
106: PetscCallMPI(MPI_Iallreduce(sendbuf, recvbuf, count, datatype, op, comm, request));
107: #else
108: PetscCall(MPIU_Allreduce(sendbuf, recvbuf, count, datatype, op, comm));
109: *request = MPI_REQUEST_NULL;
110: #endif
111: PetscFunctionReturn(PETSC_SUCCESS);
112: }
114: static PetscErrorCode KSPView_PIPELCG(KSP ksp, PetscViewer viewer)
115: {
116: KSP_CG_PIPE_L *plcg = (KSP_CG_PIPE_L *)ksp->data;
117: PetscBool iascii = PETSC_FALSE, isstring = PETSC_FALSE;
119: PetscFunctionBegin;
120: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &iascii));
121: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERSTRING, &isstring));
122: if (iascii) {
123: PetscCall(PetscViewerASCIIPrintf(viewer, " Pipeline depth: %" PetscInt_FMT "\n", plcg->l));
124: PetscCall(PetscViewerASCIIPrintf(viewer, " Minimal eigenvalue estimate %g\n", (double)plcg->lmin));
125: PetscCall(PetscViewerASCIIPrintf(viewer, " Maximal eigenvalue estimate %g\n", (double)plcg->lmax));
126: } else if (isstring) {
127: PetscCall(PetscViewerStringSPrintf(viewer, " Pipeline depth: %" PetscInt_FMT "\n", plcg->l));
128: PetscCall(PetscViewerStringSPrintf(viewer, " Minimal eigenvalue estimate %g\n", (double)plcg->lmin));
129: PetscCall(PetscViewerStringSPrintf(viewer, " Maximal eigenvalue estimate %g\n", (double)plcg->lmax));
130: }
131: PetscFunctionReturn(PETSC_SUCCESS);
132: }
134: static PetscErrorCode KSPSolve_InnerLoop_PIPELCG(KSP ksp)
135: {
136: KSP_CG_PIPE_L *plcg = (KSP_CG_PIPE_L *)ksp->data;
137: Mat A = NULL, Pmat = NULL;
138: PetscInt it = 0, max_it = ksp->max_it, l = plcg->l, i = 0, j = 0, k = 0;
139: PetscInt start = 0, middle = 0, end = 0;
140: Vec *Z = plcg->Z, *U = plcg->U, *V = plcg->V, *Q = plcg->Q;
141: Vec x = NULL, p = NULL, temp = NULL;
142: PetscScalar sum_dummy = 0.0, eta = 0.0, zeta = 0.0, lambda = 0.0;
143: PetscReal dp = 0.0, tmp = 0.0, beta = 0.0, invbeta2 = 0.0;
144: MPI_Comm comm;
146: PetscFunctionBegin;
147: x = ksp->vec_sol;
148: p = plcg->p;
150: comm = PetscObjectComm((PetscObject)ksp);
151: PetscCall(PCGetOperators(ksp->pc, &A, &Pmat));
153: for (it = 0; it < max_it + l; ++it) {
154: /* Multiplication z_{it+1} = Az_{it} */
155: /* Shift the U vector pointers */
156: temp = U[2];
157: for (i = 2; i > 0; i--) U[i] = U[i - 1];
158: U[0] = temp;
159: if (it < l) {
160: /* SpMV and Sigma-shift and Prec */
161: PetscCall(MatMult(A, Z[l - it], U[0]));
162: PetscCall(VecAXPY(U[0], -sigma(it), U[1]));
163: PetscCall(KSP_PCApply(ksp, U[0], Z[l - it - 1]));
164: if (it < l - 1) PetscCall(VecCopy(Z[l - it - 1], Q[3 * it]));
165: } else {
166: /* Shift the Z vector pointers */
167: temp = Z[PetscMax(l, 2)];
168: for (i = PetscMax(l, 2); i > 0; --i) Z[i] = Z[i - 1];
169: Z[0] = temp;
170: /* SpMV and Prec */
171: PetscCall(MatMult(A, Z[1], U[0]));
172: PetscCall(KSP_PCApply(ksp, U[0], Z[0]));
173: }
175: /* Adjust the G matrix */
176: if (it >= l) {
177: if (it == l) {
178: /* MPI_Wait for G(0,0),scale V0 and Z and U and Q vectors with 1/beta */
179: PetscCallMPI(MPI_Wait(&req(0), MPI_STATUS_IGNORE));
180: beta = PetscSqrtReal(PetscRealPart(G(0, 0)));
181: G(0, 0) = 1.0;
182: PetscCall(VecAXPY(V[0], 1.0 / beta, p)); /* this assumes V[0] to be zero initially */
183: for (j = 0; j <= PetscMax(l, 2); ++j) PetscCall(VecScale(Z[j], 1.0 / beta));
184: for (j = 0; j <= 2; ++j) PetscCall(VecScale(U[j], 1.0 / beta));
185: for (j = 0; j < l - 1; ++j) PetscCall(VecScale(Q[3 * j], 1.0 / beta));
186: }
188: /* MPI_Wait until the dot products,started l iterations ago,are completed */
189: PetscCallMPI(MPI_Wait(&req(it - l + 1), MPI_STATUS_IGNORE));
190: if (it >= 2 * l) {
191: for (j = PetscMax(0, it - 3 * l + 1); j <= it - 2 * l; j++) { G(j, it - l + 1) = G(it - 2 * l + 1, j + l); /* exploit symmetry in G matrix */ }
192: }
194: if (it <= 2 * l - 1) {
195: invbeta2 = 1.0 / (beta * beta);
196: /* Scale columns 1 up to l of G with 1/beta^2 */
197: for (j = PetscMax(it - 3 * l + 1, 0); j <= it - l + 1; ++j) G(j, it - l + 1) *= invbeta2;
198: }
200: for (j = PetscMax(it - 2 * l + 2, 0); j <= it - l; ++j) {
201: sum_dummy = 0.0;
202: for (k = PetscMax(it - 3 * l + 1, 0); k <= j - 1; ++k) sum_dummy = sum_dummy + G(k, j) * G(k, it - l + 1);
203: G(j, it - l + 1) = (G(j, it - l + 1) - sum_dummy) / G(j, j);
204: }
206: sum_dummy = 0.0;
207: for (k = PetscMax(it - 3 * l + 1, 0); k <= it - l; ++k) sum_dummy = sum_dummy + G(k, it - l + 1) * G(k, it - l + 1);
209: tmp = PetscRealPart(G(it - l + 1, it - l + 1) - sum_dummy);
210: /* Breakdown check */
211: if (tmp < 0) {
212: if (plcg->show_rstrt) PetscCall(PetscPrintf(comm, "Sqrt breakdown in iteration %" PetscInt_FMT ": sqrt argument is %e. Iteration was restarted.\n", ksp->its + 1, (double)tmp));
213: /* End hanging dot-products in the pipeline before exiting for-loop */
214: start = it - l + 2;
215: end = PetscMin(it + 1, max_it + 1); /* !warning! 'it' can actually be greater than 'max_it' */
216: for (i = start; i < end; ++i) PetscCallMPI(MPI_Wait(&req(i), MPI_STATUS_IGNORE));
217: break;
218: }
219: G(it - l + 1, it - l + 1) = PetscSqrtReal(tmp);
221: if (it < 2 * l) {
222: if (it == l) {
223: gamma(it - l) = (G(it - l, it - l + 1) + sigma(it - l) * G(it - l, it - l)) / G(it - l, it - l);
224: } else {
225: gamma(it - l) = (G(it - l, it - l + 1) + sigma(it - l) * G(it - l, it - l) - delta(it - l - 1) * G(it - l - 1, it - l)) / G(it - l, it - l);
226: }
227: delta(it - l) = G(it - l + 1, it - l + 1) / G(it - l, it - l);
228: } else {
229: gamma(it - l) = (G(it - l, it - l) * gamma(it - 2 * l) + G(it - l, it - l + 1) * delta(it - 2 * l) - G(it - l - 1, it - l) * delta(it - l - 1)) / G(it - l, it - l);
230: delta(it - l) = (G(it - l + 1, it - l + 1) * delta(it - 2 * l)) / G(it - l, it - l);
231: }
233: /* Recursively compute the next V, Q, Z and U vectors */
234: /* Shift the V vector pointers */
235: temp = V[2];
236: for (i = 2; i > 0; i--) V[i] = V[i - 1];
237: V[0] = temp;
239: /* Recurrence V vectors */
240: if (l == 1) {
241: PetscCall(VecCopy(Z[1], V[0]));
242: } else {
243: PetscCall(VecCopy(Q[0], V[0]));
244: }
245: if (it == l) {
246: PetscCall(VecAXPY(V[0], sigma(0) - gamma(it - l), V[1]));
247: } else {
248: alpha(0) = sigma(0) - gamma(it - l);
249: alpha(1) = -delta(it - l - 1);
250: PetscCall(VecMAXPY(V[0], 2, &alpha(0), &V[1]));
251: }
252: PetscCall(VecScale(V[0], 1.0 / delta(it - l)));
254: /* Recurrence Q vectors */
255: for (j = 0; j < l - 1; ++j) {
256: /* Shift the Q vector pointers */
257: temp = Q[3 * j + 2];
258: for (i = 2; i > 0; i--) Q[3 * j + i] = Q[3 * j + i - 1];
259: Q[3 * j] = temp;
261: if (j < l - 2) {
262: PetscCall(VecCopy(Q[3 * (j + 1)], Q[3 * j]));
263: } else {
264: PetscCall(VecCopy(Z[1], Q[3 * j]));
265: }
266: if (it == l) {
267: PetscCall(VecAXPY(Q[3 * j], sigma(j + 1) - gamma(it - l), Q[3 * j + 1]));
268: } else {
269: alpha(0) = sigma(j + 1) - gamma(it - l);
270: alpha(1) = -delta(it - l - 1);
271: PetscCall(VecMAXPY(Q[3 * j], 2, &alpha(0), &Q[3 * j + 1]));
272: }
273: PetscCall(VecScale(Q[3 * j], 1.0 / delta(it - l)));
274: }
276: /* Recurrence Z and U vectors */
277: if (it == l) {
278: PetscCall(VecAXPY(Z[0], -gamma(it - l), Z[1]));
279: PetscCall(VecAXPY(U[0], -gamma(it - l), U[1]));
280: } else {
281: alpha(0) = -gamma(it - l);
282: alpha(1) = -delta(it - l - 1);
283: PetscCall(VecMAXPY(Z[0], 2, &alpha(0), &Z[1]));
284: PetscCall(VecMAXPY(U[0], 2, &alpha(0), &U[1]));
285: }
286: PetscCall(VecScale(Z[0], 1.0 / delta(it - l)));
287: PetscCall(VecScale(U[0], 1.0 / delta(it - l)));
288: }
290: /* Compute and communicate the dot products */
291: if (it < l) {
292: for (j = 0; j < it + 2; ++j) { PetscCall((*U[0]->ops->dot_local)(U[0], Z[l - j], &G(j, it + 1))); /* dot-products (U[0],Z[j]) */ }
293: PetscCall(MPIPetsc_Iallreduce(MPI_IN_PLACE, &G(0, it + 1), it + 2, MPIU_SCALAR, MPIU_SUM, comm, &req(it + 1)));
294: } else if ((it >= l) && (it < max_it)) {
295: middle = it - l + 2;
296: end = it + 2;
297: PetscCall((*U[0]->ops->dot_local)(U[0], V[0], &G(it - l + 1, it + 1))); /* dot-product (U[0],V[0]) */
298: for (j = middle; j < end; ++j) { PetscCall((*U[0]->ops->dot_local)(U[0], plcg->Z[it + 1 - j], &G(j, it + 1))); /* dot-products (U[0],Z[j]) */ }
299: PetscCall(MPIPetsc_Iallreduce(MPI_IN_PLACE, &G(it - l + 1, it + 1), l + 1, MPIU_SCALAR, MPIU_SUM, comm, &req(it + 1)));
300: }
302: /* Compute solution vector and residual norm */
303: if (it >= l) {
304: if (it == l) {
305: if (ksp->its != 0) ++ksp->its;
306: eta = gamma(0);
307: zeta = beta;
308: PetscCall(VecCopy(V[1], p));
309: PetscCall(VecScale(p, 1.0 / eta));
310: PetscCall(VecAXPY(x, zeta, p));
311: dp = beta;
312: } else if (it > l) {
313: k = it - l;
314: ++ksp->its;
315: lambda = delta(k - 1) / eta;
316: eta = gamma(k) - lambda * delta(k - 1);
317: zeta = -lambda * zeta;
318: PetscCall(VecScale(p, -delta(k - 1) / eta));
319: PetscCall(VecAXPY(p, 1.0 / eta, V[1]));
320: PetscCall(VecAXPY(x, zeta, p));
321: dp = PetscAbsScalar(zeta);
322: }
323: ksp->rnorm = dp;
324: PetscCall(KSPLogResidualHistory(ksp, dp));
325: PetscCall(KSPMonitor(ksp, ksp->its, dp));
326: PetscCall((*ksp->converged)(ksp, ksp->its, dp, &ksp->reason, ksp->cnvP));
328: if (ksp->its >= max_it && !ksp->reason) ksp->reason = KSP_DIVERGED_ITS;
329: if (ksp->reason) {
330: /* End hanging dot-products in the pipeline before exiting for-loop */
331: start = it - l + 2;
332: end = PetscMin(it + 2, max_it + 1); /* !warning! 'it' can actually be greater than 'max_it' */
333: for (i = start; i < end; ++i) PetscCallMPI(MPI_Wait(&req(i), MPI_STATUS_IGNORE));
334: break;
335: }
336: }
337: } /* End inner for loop */
338: PetscFunctionReturn(PETSC_SUCCESS);
339: }
341: static PetscErrorCode KSPSolve_ReInitData_PIPELCG(KSP ksp)
342: {
343: KSP_CG_PIPE_L *plcg = (KSP_CG_PIPE_L *)ksp->data;
344: PetscInt i = 0, j = 0, l = plcg->l, max_it = ksp->max_it;
346: PetscFunctionBegin;
347: for (i = 0; i < PetscMax(3, l + 1); ++i) PetscCall(VecSet(plcg->Z[i], 0.0));
348: for (i = 1; i < 3; ++i) PetscCall(VecSet(plcg->U[i], 0.0));
349: for (i = 0; i < 3; ++i) PetscCall(VecSet(plcg->V[i], 0.0));
350: for (i = 0; i < 3 * (l - 1) + 1; ++i) PetscCall(VecSet(plcg->Q[i], 0.0));
351: for (j = 0; j < (max_it + 1); ++j) {
352: gamma(j) = 0.0;
353: delta(j) = 0.0;
354: for (i = 0; i < (2 * l + 1); ++i) G_noshift(i, j) = 0.0;
355: }
356: PetscFunctionReturn(PETSC_SUCCESS);
357: }
359: /*
360: KSPSolve_PIPELCG - This routine actually applies the pipelined(l) conjugate gradient method
361: */
362: static PetscErrorCode KSPSolve_PIPELCG(KSP ksp)
363: {
364: KSP_CG_PIPE_L *plcg = (KSP_CG_PIPE_L *)ksp->data;
365: Mat A = NULL, Pmat = NULL;
366: Vec b = NULL, x = NULL, p = NULL;
367: PetscInt max_it = ksp->max_it, l = plcg->l;
368: PetscInt i = 0, outer_it = 0, curr_guess_zero = 0;
369: PetscReal lmin = plcg->lmin, lmax = plcg->lmax;
370: PetscBool diagonalscale = PETSC_FALSE;
371: MPI_Comm comm;
373: PetscFunctionBegin;
374: comm = PetscObjectComm((PetscObject)ksp);
375: PetscCall(PCGetDiagonalScale(ksp->pc, &diagonalscale));
376: PetscCheck(!diagonalscale, comm, PETSC_ERR_SUP, "Krylov method %s does not support diagonal scaling", ((PetscObject)ksp)->type_name);
378: x = ksp->vec_sol;
379: b = ksp->vec_rhs;
380: p = plcg->p;
382: PetscCall(PetscCalloc1((max_it + 1) * (2 * l + 1), &plcg->G));
383: PetscCall(PetscCalloc1(max_it + 1, &plcg->gamma));
384: PetscCall(PetscCalloc1(max_it + 1, &plcg->delta));
385: PetscCall(PetscCalloc1(max_it + 1, &plcg->req));
387: PetscCall(PCGetOperators(ksp->pc, &A, &Pmat));
389: for (i = 0; i < l; ++i) sigma(i) = (0.5 * (lmin + lmax) + (0.5 * (lmax - lmin) * PetscCosReal(PETSC_PI * (2.0 * i + 1.0) / (2.0 * l))));
391: ksp->its = 0;
392: outer_it = 0;
393: curr_guess_zero = !!ksp->guess_zero;
395: while (ksp->its < max_it) { /* OUTER LOOP (gmres-like restart to handle breakdowns) */
396: /* RESTART LOOP */
397: if (!curr_guess_zero) {
398: PetscCall(KSP_MatMult(ksp, A, x, plcg->U[0])); /* u <- b - Ax */
399: PetscCall(VecAYPX(plcg->U[0], -1.0, b));
400: } else {
401: PetscCall(VecCopy(b, plcg->U[0])); /* u <- b (x is 0) */
402: }
403: PetscCall(KSP_PCApply(ksp, plcg->U[0], p)); /* p <- Bu */
405: if (outer_it > 0) {
406: /* Re-initialize Z,U,V,Q,gamma,delta,G after restart occurred */
407: PetscCall(KSPSolve_ReInitData_PIPELCG(ksp));
408: }
410: PetscCall((*plcg->U[0]->ops->dot_local)(plcg->U[0], p, &G(0, 0)));
411: PetscCall(MPIPetsc_Iallreduce(MPI_IN_PLACE, &G(0, 0), 1, MPIU_SCALAR, MPIU_SUM, comm, &req(0)));
412: PetscCall(VecCopy(p, plcg->Z[l]));
414: PetscCall(KSPSolve_InnerLoop_PIPELCG(ksp));
416: if (ksp->reason) break; /* convergence or divergence */
417: ++outer_it;
418: curr_guess_zero = 0;
419: }
421: if (!ksp->reason) ksp->reason = KSP_DIVERGED_ITS;
422: PetscCall(PetscFree(plcg->G));
423: PetscCall(PetscFree(plcg->gamma));
424: PetscCall(PetscFree(plcg->delta));
425: PetscCall(PetscFree(plcg->req));
426: PetscFunctionReturn(PETSC_SUCCESS);
427: }
429: /*MC
430: KSPPIPELCG - Deep pipelined (length l) Conjugate Gradient method {cite}`cornelis2018communication` and {cite}`cools2019numerically`.
431: This method has only a single non-blocking global
432: reduction per iteration, compared to 2 blocking reductions for standard `KSPCG`. The reduction is overlapped by the
433: matrix-vector product and preconditioner application of the next l iterations. The pipeline length l is a parameter
434: of the method. [](sec_pipelineksp)
436: Options Database Keys:
437: + -ksp_pipelcg_pipel - pipelined length
438: . -ksp_pipelcg_lmin - approximation to the smallest eigenvalue of the preconditioned operator (default: 0.0)
439: . -ksp_pipelcg_lmax - approximation to the largest eigenvalue of the preconditioned operator (default: 0.0)
440: - -ksp_pipelcg_monitor - output where/why the method restarts when a sqrt breakdown occurs
442: Example usage:
443: .vb
444: KSP tutorials ex2, no preconditioner, pipel = 2, lmin = 0.0, lmax = 8.0 :
445: $mpiexec -n 14 ./ex2 -m 1000 -n 1000 -ksp_type pipelcg -pc_type none -ksp_norm_type natural
446: -ksp_rtol 1e-10 -ksp_max_it 1000 -ksp_pipelcg_pipel 2 -ksp_pipelcg_lmin 0.0 -ksp_pipelcg_lmax 8.0 -log_view
447: SNES tutorials ex48, bjacobi preconditioner, pipel = 3, lmin = 0.0, lmax = 2.0, show restart information :
448: $mpiexec -n 14 ./ex48 -M 150 -P 100 -ksp_type pipelcg -pc_type bjacobi -ksp_rtol 1e-10 -ksp_pipelcg_pipel 3
449: -ksp_pipelcg_lmin 0.0 -ksp_pipelcg_lmax 2.0 -ksp_pipelcg_monitor -log_view
450: .ve
452: Level: advanced
454: Notes:
455: MPI configuration may be necessary for reductions to make asynchronous progress, which is important for
456: performance of pipelined methods. See [](doc_faq_pipelined)
458: Contributed by:
459: Siegfried Cools, University of Antwerp, Dept. Mathematics and Computer Science,
460: funded by Flemish Research Foundation (FWO) grant number 12H4617N.
462: .seealso: [](ch_ksp), [](sec_pipelineksp), [](doc_faq_pipelined), `KSPCreate()`, `KSPSetType()`, `KSPType`, `KSPCG`, `KSPPIPECG`, `KSPPIPECGRR`, `KSPPGMRES`,
463: `KSPPIPEBCGS`, `KSPSetPCSide()`, `KSPGROPPCG`
464: M*/
465: PETSC_EXTERN PetscErrorCode KSPCreate_PIPELCG(KSP ksp)
466: {
467: KSP_CG_PIPE_L *plcg = NULL;
469: PetscFunctionBegin;
470: PetscCall(PetscNew(&plcg));
471: ksp->data = (void *)plcg;
473: PetscCall(KSPSetSupportedNorm(ksp, KSP_NORM_NONE, PC_LEFT, 1));
474: PetscCall(KSPSetSupportedNorm(ksp, KSP_NORM_NATURAL, PC_LEFT, 2));
476: ksp->ops->setup = KSPSetUp_PIPELCG;
477: ksp->ops->solve = KSPSolve_PIPELCG;
478: ksp->ops->reset = KSPReset_PIPELCG;
479: ksp->ops->destroy = KSPDestroy_PIPELCG;
480: ksp->ops->view = KSPView_PIPELCG;
481: ksp->ops->setfromoptions = KSPSetFromOptions_PIPELCG;
482: ksp->ops->buildsolution = KSPBuildSolutionDefault;
483: ksp->ops->buildresidual = KSPBuildResidualDefault;
484: PetscFunctionReturn(PETSC_SUCCESS);
485: }