1: !
  2: !
  3: !  Description: Builds a parallel vector with 1 component on the first
  4: !               processor, 2 on the second, etc.  Then each processor adds
  5: !               one to all elements except the last rank.
  6: !
  7: !/*T
  8: !   Concepts: vectors^assembling
  9: !   Processors: n
 10: !T*/
 11: ! -----------------------------------------------------------------------

 13:       program main
 14: #include <petsc/finclude/petscvec.h>
 15:       use petscvec
 16:       implicit none

 18: ! - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
 19: !                 Beginning of program
 20: ! - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

 22:       Vec     x
 23:       PetscInt N,i,ione
 24:       PetscErrorCode ierr
 25:       PetscMPIInt rank
 26:       PetscScalar  one

 28:       call PetscInitialize(PETSC_NULL_CHARACTER,ierr)
 29:       if (ierr .ne. 0) then
 30:         print*,'PetscInitialize failed'
 31:         stop
 32:       endif
 33:       one   = 1.0
 34:       call MPI_Comm_rank(PETSC_COMM_WORLD,rank,ierr)

 36: !  Create a parallel vector.
 37: !   - In this case, we specify the size of the local portion on
 38: !     each processor, and PETSc computes the global size.  Alternatively,
 39: !     if we pass the global size and use PETSC_DECIDE for the
 40: !     local size PETSc will choose a reasonable partition trying
 41: !     to put nearly an equal number of elements on each processor.

 43:       N = rank + 1
 44:       call VecCreateMPI(PETSC_COMM_WORLD,N,PETSC_DECIDE,x,ierr)
 45:       call VecGetSize(x,N,ierr)
 46:       call VecSet(x,one,ierr)

 48: !  Set the vector elements.
 49: !   - Note that VecSetValues() uses 0-based row and column numbers
 50: !     in Fortran as well as in C.
 51: !   - Always specify global locations of vector entries.
 52: !   - Each processor can contribute any vector entries,
 53: !     regardless of which processor "owns" them; any nonlocal
 54: !     contributions will be transferred to the appropriate processor
 55: !     during the assembly process.
 56: !   - In this example, the flag ADD_VALUES indicates that all
 57: !     contributions will be added together.

 59:       ione = 1
 60:       do 100 i=0,N-rank-1
 61:          call VecSetValues(x,ione,i,one,ADD_VALUES,ierr)
 62:  100  continue

 64: !  Assemble vector, using the 2-step process:
 65: !    VecAssemblyBegin(), VecAssemblyEnd()
 66: !  Computations can be done while messages are in transition
 67: !  by placing code between these two statements.

 69:       call VecAssemblyBegin(x,ierr)
 70:       call VecAssemblyEnd(x,ierr)

 72: !     Test VecGetValues() with scalar entries
 73:       if (rank .eq. 0) then
 74:         ione = 0
 75:         call VecGetValues(x,ione,i,one,ierr)
 76:       endif

 78: !  View the vector; then destroy it.

 80:       call VecView(x,PETSC_VIEWER_STDOUT_WORLD,ierr)
 81:       call VecDestroy(x,ierr)

 83:       call PetscFinalize(ierr)
 84:       end

 86: !/*TEST
 87: !
 88: !     test:
 89: !       nsize: 2
 90: !       filter: grep -v "MPI processes"
 91: !
 92: !TEST*/