Actual source code: ex6f.F90
1: !
2: ! Description: This example demonstrates repeated linear solves as
3: ! well as the use of different preconditioner and linear system
4: ! matrices. This example also illustrates how to save PETSc objects
5: ! in common blocks.
6: !
7: !
9: program main
10: #include <petsc/finclude/petscksp.h>
11: use petscksp
12: implicit none
14: ! Variables:
15: !
16: ! A - matrix that defines linear system
17: ! ksp - KSP context
18: ! ksp - KSP context
19: ! x, b, u - approx solution, RHS, exact solution vectors
20: !
21: Vec x,u,b
22: Mat A,A2
23: KSP ksp
24: PetscInt i,j,II,JJ,m,n
25: PetscInt Istart,Iend
26: PetscInt nsteps,one
27: PetscErrorCode ierr
28: PetscBool flg
29: PetscScalar v
31: PetscCallA(PetscInitialize(ierr))
32: m = 3
33: n = 3
34: nsteps = 2
35: one = 1
36: PetscCallA(PetscOptionsGetInt(PETSC_NULL_OPTIONS,PETSC_NULL_CHARACTER,'-m',m,flg,ierr))
37: PetscCallA(PetscOptionsGetInt(PETSC_NULL_OPTIONS,PETSC_NULL_CHARACTER,'-n',n,flg,ierr))
38: PetscCallA(PetscOptionsGetInt(PETSC_NULL_OPTIONS,PETSC_NULL_CHARACTER,'-nsteps',nsteps,flg,ierr))
40: ! Create parallel matrix, specifying only its global dimensions.
41: ! When using MatCreate(), the matrix format can be specified at
42: ! runtime. Also, the parallel partitioning of the matrix is
43: ! determined by PETSc at runtime.
45: PetscCallA(MatCreate(PETSC_COMM_WORLD,A,ierr))
46: PetscCallA(MatSetSizes(A,PETSC_DECIDE,PETSC_DECIDE,m*n,m*n,ierr))
47: PetscCallA(MatSetFromOptions(A,ierr))
48: PetscCallA(MatSetUp(A,ierr))
50: ! The matrix is partitioned by contiguous chunks of rows across the
51: ! processors. Determine which rows of the matrix are locally owned.
53: PetscCallA(MatGetOwnershipRange(A,Istart,Iend,ierr))
55: ! Set matrix elements.
56: ! - Each processor needs to insert only elements that it owns
57: ! locally (but any non-local elements will be sent to the
58: ! appropriate processor during matrix assembly).
59: ! - Always specify global rows and columns of matrix entries.
61: do 10, II=Istart,Iend-1
62: v = -1.0
63: i = II/n
64: j = II - i*n
65: if (i.gt.0) then
66: JJ = II - n
67: PetscCallA(MatSetValues(A,one,[II],one,[JJ],[v],ADD_VALUES,ierr))
68: endif
69: if (i.lt.m-1) then
70: JJ = II + n
71: PetscCallA(MatSetValues(A,one,[II],one,[JJ],[v],ADD_VALUES,ierr))
72: endif
73: if (j.gt.0) then
74: JJ = II - 1
75: PetscCallA(MatSetValues(A,one,[II],one,[JJ],[v],ADD_VALUES,ierr))
76: endif
77: if (j.lt.n-1) then
78: JJ = II + 1
79: PetscCallA(MatSetValues(A,one,[II],one,[JJ],[v],ADD_VALUES,ierr))
80: endif
81: v = 4.0
82: PetscCallA( MatSetValues(A,one,[II],one,[II],[v],ADD_VALUES,ierr))
83: 10 continue
85: ! Assemble matrix, using the 2-step process:
86: ! MatAssemblyBegin(), MatAssemblyEnd()
87: ! Computations can be done while messages are in transition
88: ! by placing code between these two statements.
90: PetscCallA(MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY,ierr))
91: PetscCallA(MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY,ierr))
93: ! Create parallel vectors.
94: ! - When using VecCreate(), the parallel partitioning of the vector
95: ! is determined by PETSc at runtime.
96: ! - Note: We form 1 vector from scratch and then duplicate as needed.
98: PetscCallA(VecCreate(PETSC_COMM_WORLD,u,ierr))
99: PetscCallA(VecSetSizes(u,PETSC_DECIDE,m*n,ierr))
100: PetscCallA(VecSetFromOptions(u,ierr))
101: PetscCallA(VecDuplicate(u,b,ierr))
102: PetscCallA(VecDuplicate(b,x,ierr))
104: ! Create linear solver context
106: PetscCallA(KSPCreate(PETSC_COMM_WORLD,ksp,ierr))
108: ! Set runtime options (e.g., -ksp_type <type> -pc_type <type>)
110: PetscCallA(KSPSetFromOptions(ksp,ierr))
112: ! Solve several linear systems in succession
114: do 100 i=1,nsteps
115: PetscCallA(solve1(ksp,A,x,b,u,i,nsteps,A2,ierr))
116: 100 continue
118: ! Free work space. All PETSc objects should be destroyed when they
119: ! are no longer needed.
121: PetscCallA(VecDestroy(u,ierr))
122: PetscCallA(VecDestroy(x,ierr))
123: PetscCallA(VecDestroy(b,ierr))
124: PetscCallA(MatDestroy(A,ierr))
125: PetscCallA(KSPDestroy(ksp,ierr))
127: PetscCallA(PetscFinalize(ierr))
128: end
130: ! -----------------------------------------------------------------------
131: !
132: subroutine solve1(ksp,A,x,b,u,count,nsteps,A2,ierr)
133: use petscksp
134: implicit none
136: !
137: ! solve1 - This routine is used for repeated linear system solves.
138: ! We update the linear system matrix each time, but retain the same
139: ! preconditioning matrix for all linear solves.
140: !
141: ! A - linear system matrix
142: ! A2 - preconditioning matrix
143: !
144: PetscScalar v,val
145: PetscInt II,Istart,Iend
146: PetscInt count,nsteps,one
147: PetscErrorCode ierr
148: Mat A
149: KSP ksp
150: Vec x,b,u
152: ! Use common block to retain matrix between successive subroutine calls
153: Mat A2
154: PetscMPIInt rank
155: PetscBool pflag
156: common /my_data/ pflag,rank
158: one = 1
159: ! First time thorough: Create new matrix to define the linear system
160: if (count .eq. 1) then
161: PetscCallMPIA(MPI_Comm_rank(PETSC_COMM_WORLD,rank,ierr))
162: pflag = .false.
163: PetscCallA(PetscOptionsHasName(PETSC_NULL_OPTIONS,PETSC_NULL_CHARACTER,'-mat_view',pflag,ierr))
164: if (pflag) then
165: if (rank .eq. 0) write(6,100)
166: call PetscFlush(6)
167: endif
168: PetscCallA(MatConvert(A,MATSAME,MAT_INITIAL_MATRIX,A2,ierr))
169: ! All other times: Set previous solution as initial guess for next solve.
170: else
171: PetscCallA(KSPSetInitialGuessNonzero(ksp,PETSC_TRUE,ierr))
172: endif
174: ! Alter the matrix A a bit
175: PetscCallA(MatGetOwnershipRange(A,Istart,Iend,ierr))
176: do 20, II=Istart,Iend-1
177: v = 2.0
178: PetscCallA(MatSetValues(A,one,[II],one,[II],[v],ADD_VALUES,ierr))
179: 20 continue
180: PetscCallA(MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY,ierr))
181: if (pflag) then
182: if (rank .eq. 0) write(6,110)
183: call PetscFlush(6)
184: endif
185: PetscCallA(MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY,ierr))
187: ! Set the exact solution; compute the right-hand-side vector
188: val = 1.0*real(count)
189: PetscCallA(VecSet(u,val,ierr))
190: PetscCallA(MatMult(A,u,b,ierr))
192: ! Set operators, keeping the identical preconditioner matrix for
193: ! all linear solves. This approach is often effective when the
194: ! linear systems do not change very much between successive steps.
195: PetscCallA(KSPSetReusePreconditioner(ksp,PETSC_TRUE,ierr))
196: PetscCallA(KSPSetOperators(ksp,A,A2,ierr))
198: ! Solve linear system
199: PetscCallA(KSPSolve(ksp,b,x,ierr))
201: ! Destroy the preconditioner matrix on the last time through
202: if (count .eq. nsteps) PetscCallA(MatDestroy(A2,ierr))
204: 100 format('previous matrix: preconditioning')
205: 110 format('next matrix: defines linear system')
207: end
209: !/*TEST
210: !
211: ! test:
212: ! args: -pc_type jacobi -mat_view -ksp_monitor_short -ksp_gmres_cgs_refinement_type refine_always
213: !
214: !TEST*/