Actual source code: pipecr.c
1: #include <petsc/private/kspimpl.h>
3: /*
4: KSPSetUp_PIPECR - Sets up the workspace needed by the PIPECR method.
6: This is called once, usually automatically by KSPSolve() or KSPSetUp()
7: but can be called directly by KSPSetUp()
8: */
9: static PetscErrorCode KSPSetUp_PIPECR(KSP ksp)
10: {
11: PetscFunctionBegin;
12: /* get work vectors needed by PIPECR */
13: PetscCall(KSPSetWorkVecs(ksp, 7));
14: PetscFunctionReturn(PETSC_SUCCESS);
15: }
17: /*
18: KSPSolve_PIPECR - This routine actually applies the pipelined conjugate residual method
19: */
20: static PetscErrorCode KSPSolve_PIPECR(KSP ksp)
21: {
22: PetscInt i;
23: PetscScalar alpha = 0.0, beta = 0.0, gamma, gammaold = 0.0, delta;
24: PetscReal dp = 0.0;
25: Vec X, B, Z, P, W, Q, U, M, N;
26: Mat Amat, Pmat;
28: PetscFunctionBegin;
29: X = ksp->vec_sol;
30: B = ksp->vec_rhs;
31: M = ksp->work[0];
32: Z = ksp->work[1];
33: P = ksp->work[2];
34: N = ksp->work[3];
35: W = ksp->work[4];
36: Q = ksp->work[5];
37: U = ksp->work[6];
39: PetscCall(PCGetOperators(ksp->pc, &Amat, &Pmat));
41: ksp->its = 0;
42: /* we don't have an R vector, so put the (unpreconditioned) residual in w for now */
43: if (!ksp->guess_zero) {
44: PetscCall(KSP_MatMult(ksp, Amat, X, W)); /* w <- b - Ax */
45: PetscCall(VecAYPX(W, -1.0, B));
46: } else {
47: PetscCall(VecCopy(B, W)); /* w <- b (x is 0) */
48: }
49: PetscCall(KSP_PCApply(ksp, W, U)); /* u <- Bw */
51: switch (ksp->normtype) {
52: case KSP_NORM_PRECONDITIONED:
53: PetscCall(VecNormBegin(U, NORM_2, &dp)); /* dp <- u'*u = e'*A'*B'*B*A'*e' */
54: PetscCall(PetscCommSplitReductionBegin(PetscObjectComm((PetscObject)U)));
55: PetscCall(KSP_MatMult(ksp, Amat, U, W)); /* w <- Au */
56: PetscCall(VecNormEnd(U, NORM_2, &dp));
57: break;
58: case KSP_NORM_NONE:
59: PetscCall(KSP_MatMult(ksp, Amat, U, W));
60: dp = 0.0;
61: break;
62: default:
63: SETERRQ(PetscObjectComm((PetscObject)ksp), PETSC_ERR_SUP, "%s", KSPNormTypes[ksp->normtype]);
64: }
65: PetscCall(KSPLogResidualHistory(ksp, dp));
66: PetscCall(KSPMonitor(ksp, 0, dp));
67: ksp->rnorm = dp;
68: PetscCall((*ksp->converged)(ksp, 0, dp, &ksp->reason, ksp->cnvP)); /* test for convergence */
69: if (ksp->reason) PetscFunctionReturn(PETSC_SUCCESS);
71: i = 0;
72: do {
73: PetscCall(KSP_PCApply(ksp, W, M)); /* m <- Bw */
75: if (i > 0 && ksp->normtype == KSP_NORM_PRECONDITIONED) PetscCall(VecNormBegin(U, NORM_2, &dp));
76: PetscCall(VecDotBegin(W, U, &gamma));
77: PetscCall(VecDotBegin(M, W, &delta));
78: PetscCall(PetscCommSplitReductionBegin(PetscObjectComm((PetscObject)U)));
80: PetscCall(KSP_MatMult(ksp, Amat, M, N)); /* n <- Am */
82: if (i > 0 && ksp->normtype == KSP_NORM_PRECONDITIONED) PetscCall(VecNormEnd(U, NORM_2, &dp));
83: PetscCall(VecDotEnd(W, U, &gamma));
84: PetscCall(VecDotEnd(M, W, &delta));
86: if (i > 0) {
87: if (ksp->normtype == KSP_NORM_NONE) dp = 0.0;
88: ksp->rnorm = dp;
89: PetscCall(KSPLogResidualHistory(ksp, dp));
90: PetscCall(KSPMonitor(ksp, i, dp));
91: PetscCall((*ksp->converged)(ksp, i, dp, &ksp->reason, ksp->cnvP));
92: if (ksp->reason) PetscFunctionReturn(PETSC_SUCCESS);
93: }
95: if (i == 0) {
96: alpha = gamma / delta;
97: PetscCall(VecCopy(N, Z)); /* z <- n */
98: PetscCall(VecCopy(M, Q)); /* q <- m */
99: PetscCall(VecCopy(U, P)); /* p <- u */
100: } else {
101: beta = gamma / gammaold;
102: alpha = gamma / (delta - beta / alpha * gamma);
103: PetscCall(VecAYPX(Z, beta, N)); /* z <- n + beta * z */
104: PetscCall(VecAYPX(Q, beta, M)); /* q <- m + beta * q */
105: PetscCall(VecAYPX(P, beta, U)); /* p <- u + beta * p */
106: }
107: PetscCall(VecAXPY(X, alpha, P)); /* x <- x + alpha * p */
108: PetscCall(VecAXPY(U, -alpha, Q)); /* u <- u - alpha * q */
109: PetscCall(VecAXPY(W, -alpha, Z)); /* w <- w - alpha * z */
110: gammaold = gamma;
111: i++;
112: ksp->its = i;
114: /* if (i%50 == 0) { */
115: /* PetscCall(KSP_MatMult(ksp,Amat,X,W)); /\* w <- b - Ax *\/ */
116: /* PetscCall(VecAYPX(W,-1.0,B)); */
117: /* PetscCall(KSP_PCApply(ksp,W,U)); */
118: /* PetscCall(KSP_MatMult(ksp,Amat,U,W)); */
119: /* } */
121: } while (i <= ksp->max_it);
122: if (i >= ksp->max_it) ksp->reason = KSP_DIVERGED_ITS;
123: PetscFunctionReturn(PETSC_SUCCESS);
124: }
126: /*MC
127: KSPPIPECR - Pipelined conjugate residual method {cite}`ghyselsvanroose2014`. [](sec_pipelineksp)
129: Level: intermediate
131: Notes:
132: This method has only a single non-blocking reduction per iteration, compared to 2 for standard `KSPCR`. The
133: non-blocking reduction is overlapped by the matrix-vector product, but not the preconditioner application.
135: See also `KSPPIPECG`, where the reduction is overlapped with the matrix-vector product.
137: MPI configuration may be necessary for reductions to make asynchronous progress, which is important for performance of pipelined methods.
138: See [](doc_faq_pipelined)
140: Contributed by:
141: Pieter Ghysels, Universiteit Antwerpen, Intel Exascience lab Flanders
143: .seealso: [](ch_ksp), [](sec_pipelineksp), [](doc_faq_pipelined), `KSPCreate()`, `KSPSetType()`, `KSPPIPECG`, `KSPGROPPCG`, `KSPPGMRES`, `KSPCG`, `KSPCGUseSingleReduction()`
144: M*/
146: PETSC_EXTERN PetscErrorCode KSPCreate_PIPECR(KSP ksp)
147: {
148: PetscFunctionBegin;
149: PetscCall(KSPSetSupportedNorm(ksp, KSP_NORM_PRECONDITIONED, PC_LEFT, 2));
150: PetscCall(KSPSetSupportedNorm(ksp, KSP_NORM_NONE, PC_LEFT, 1));
152: ksp->ops->setup = KSPSetUp_PIPECR;
153: ksp->ops->solve = KSPSolve_PIPECR;
154: ksp->ops->destroy = KSPDestroyDefault;
155: ksp->ops->view = NULL;
156: ksp->ops->setfromoptions = NULL;
157: ksp->ops->buildsolution = KSPBuildSolutionDefault;
158: ksp->ops->buildresidual = KSPBuildResidualDefault;
159: PetscFunctionReturn(PETSC_SUCCESS);
160: }