xref: /petsc/src/ksp/ksp/impls/cr/pipecr/pipecr.c (revision 613ce9fe8f5e2bcdf7c72d914e9769b5d960fb4c)
1 #include <petsc/private/kspimpl.h>
2 
3 /*
4      KSPSetUp_PIPECR - Sets up the workspace needed by the PIPECR method.
5 
6       This is called once, usually automatically by KSPSolve() or KSPSetUp()
7      but can be called directly by KSPSetUp()
8 */
KSPSetUp_PIPECR(KSP ksp)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 }
16 
17 /*
18  KSPSolve_PIPECR - This routine actually applies the pipelined conjugate residual method
19 */
KSPSolve_PIPECR(KSP ksp)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;
27   PetscBool   diagonalscale;
28 
29   PetscFunctionBegin;
30   PetscCall(PCGetDiagonalScale(ksp->pc, &diagonalscale));
31   PetscCheck(!diagonalscale, PetscObjectComm((PetscObject)ksp), PETSC_ERR_SUP, "Krylov method %s does not support diagonal scaling", ((PetscObject)ksp)->type_name);
32 
33   X = ksp->vec_sol;
34   B = ksp->vec_rhs;
35   M = ksp->work[0];
36   Z = ksp->work[1];
37   P = ksp->work[2];
38   N = ksp->work[3];
39   W = ksp->work[4];
40   Q = ksp->work[5];
41   U = ksp->work[6];
42 
43   PetscCall(PCGetOperators(ksp->pc, &Amat, &Pmat));
44 
45   ksp->its = 0;
46   /* we don't have an R vector, so put the (unpreconditioned) residual in w for now */
47   if (!ksp->guess_zero) {
48     PetscCall(KSP_MatMult(ksp, Amat, X, W)); /*     w <- b - Ax     */
49     PetscCall(VecAYPX(W, -1.0, B));
50   } else {
51     PetscCall(VecCopy(B, W)); /*     w <- b (x is 0) */
52   }
53   PetscCall(KSP_PCApply(ksp, W, U)); /*     u <- Bw   */
54 
55   switch (ksp->normtype) {
56   case KSP_NORM_PRECONDITIONED:
57     PetscCall(VecNormBegin(U, NORM_2, &dp)); /*     dp <- u'*u = e'*A'*B'*B*A'*e'     */
58     PetscCall(PetscCommSplitReductionBegin(PetscObjectComm((PetscObject)U)));
59     PetscCall(KSP_MatMult(ksp, Amat, U, W)); /*     w <- Au   */
60     PetscCall(VecNormEnd(U, NORM_2, &dp));
61     break;
62   case KSP_NORM_NONE:
63     PetscCall(KSP_MatMult(ksp, Amat, U, W));
64     dp = 0.0;
65     break;
66   default:
67     SETERRQ(PetscObjectComm((PetscObject)ksp), PETSC_ERR_SUP, "%s", KSPNormTypes[ksp->normtype]);
68   }
69   PetscCall(KSPLogResidualHistory(ksp, dp));
70   PetscCall(KSPMonitor(ksp, 0, dp));
71   ksp->rnorm = dp;
72   PetscCall((*ksp->converged)(ksp, 0, dp, &ksp->reason, ksp->cnvP)); /* test for convergence */
73   if (ksp->reason) PetscFunctionReturn(PETSC_SUCCESS);
74 
75   i = 0;
76   do {
77     PetscCall(KSP_PCApply(ksp, W, M)); /*   m <- Bw       */
78 
79     if (i > 0 && ksp->normtype == KSP_NORM_PRECONDITIONED) PetscCall(VecNormBegin(U, NORM_2, &dp));
80     PetscCall(VecDotBegin(W, U, &gamma));
81     PetscCall(VecDotBegin(M, W, &delta));
82     PetscCall(PetscCommSplitReductionBegin(PetscObjectComm((PetscObject)U)));
83 
84     PetscCall(KSP_MatMult(ksp, Amat, M, N)); /*   n <- Am       */
85 
86     if (i > 0 && ksp->normtype == KSP_NORM_PRECONDITIONED) PetscCall(VecNormEnd(U, NORM_2, &dp));
87     PetscCall(VecDotEnd(W, U, &gamma));
88     PetscCall(VecDotEnd(M, W, &delta));
89 
90     if (i > 0) {
91       if (ksp->normtype == KSP_NORM_NONE) dp = 0.0;
92       ksp->rnorm = dp;
93       PetscCall(KSPLogResidualHistory(ksp, dp));
94       PetscCall(KSPMonitor(ksp, i, dp));
95       PetscCall((*ksp->converged)(ksp, i, dp, &ksp->reason, ksp->cnvP));
96       if (ksp->reason) PetscFunctionReturn(PETSC_SUCCESS);
97     }
98 
99     if (i == 0) {
100       alpha = gamma / delta;
101       PetscCall(VecCopy(N, Z)); /*     z <- n          */
102       PetscCall(VecCopy(M, Q)); /*     q <- m          */
103       PetscCall(VecCopy(U, P)); /*     p <- u          */
104     } else {
105       beta  = gamma / gammaold;
106       alpha = gamma / (delta - beta / alpha * gamma);
107       PetscCall(VecAYPX(Z, beta, N)); /*     z <- n + beta * z   */
108       PetscCall(VecAYPX(Q, beta, M)); /*     q <- m + beta * q   */
109       PetscCall(VecAYPX(P, beta, U)); /*     p <- u + beta * p   */
110     }
111     PetscCall(VecAXPY(X, alpha, P));  /*     x <- x + alpha * p   */
112     PetscCall(VecAXPY(U, -alpha, Q)); /*     u <- u - alpha * q   */
113     PetscCall(VecAXPY(W, -alpha, Z)); /*     w <- w - alpha * z   */
114     gammaold = gamma;
115     i++;
116     ksp->its = i;
117 
118     /* if (i%50 == 0) { */
119     /*   PetscCall(KSP_MatMult(ksp,Amat,X,W));            /\*     w <- b - Ax     *\/ */
120     /*   PetscCall(VecAYPX(W,-1.0,B)); */
121     /*   PetscCall(KSP_PCApply(ksp,W,U)); */
122     /*   PetscCall(KSP_MatMult(ksp,Amat,U,W)); */
123     /* } */
124 
125   } while (i <= ksp->max_it);
126   if (i >= ksp->max_it) ksp->reason = KSP_DIVERGED_ITS;
127   PetscFunctionReturn(PETSC_SUCCESS);
128 }
129 
130 /*MC
131    KSPPIPECR - Pipelined conjugate residual method {cite}`ghyselsvanroose2014`. [](sec_pipelineksp)
132 
133    Level: intermediate
134 
135    Notes:
136    This method has only a single non-blocking reduction per iteration, compared to 2 for standard `KSPCR`.  The
137    non-blocking reduction is overlapped by the matrix-vector product, but not the preconditioner application.
138 
139    See also `KSPPIPECG`, where the reduction is overlapped with the matrix-vector product.
140 
141    MPI configuration may be necessary for reductions to make asynchronous progress, which is important for performance of pipelined methods.
142    See [](doc_faq_pipelined)
143 
144    Contributed by:
145    Pieter Ghysels, Universiteit Antwerpen, Intel Exascience lab Flanders
146 
147 .seealso: [](ch_ksp), [](sec_pipelineksp), [](doc_faq_pipelined), `KSPCreate()`, `KSPSetType()`, `KSPPIPECG`, `KSPGROPPCG`, `KSPPGMRES`, `KSPCG`, `KSPCGUseSingleReduction()`
148 M*/
149 
KSPCreate_PIPECR(KSP ksp)150 PETSC_EXTERN PetscErrorCode KSPCreate_PIPECR(KSP ksp)
151 {
152   PetscFunctionBegin;
153   PetscCall(KSPSetSupportedNorm(ksp, KSP_NORM_PRECONDITIONED, PC_LEFT, 2));
154   PetscCall(KSPSetSupportedNorm(ksp, KSP_NORM_NONE, PC_LEFT, 1));
155 
156   ksp->ops->setup          = KSPSetUp_PIPECR;
157   ksp->ops->solve          = KSPSolve_PIPECR;
158   ksp->ops->destroy        = KSPDestroyDefault;
159   ksp->ops->view           = NULL;
160   ksp->ops->setfromoptions = NULL;
161   ksp->ops->buildsolution  = KSPBuildSolutionDefault;
162   ksp->ops->buildresidual  = KSPBuildResidualDefault;
163   PetscFunctionReturn(PETSC_SUCCESS);
164 }
165