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! ----------------------------------------------------------------------- 8 9program main 10#include <petsc/finclude/petscvec.h> 11 use petscvec 12 implicit none 13 14! - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 15! Beginning of program 16! - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 17 18 Vec x 19 PetscInt N, i, ione 20 PetscErrorCode ierr 21 PetscMPIInt rank 22 PetscScalar one, value(1) 23 24 PetscCallA(PetscInitialize(ierr)) 25 one = 1.0 26 PetscCallMPIA(MPI_Comm_rank(PETSC_COMM_WORLD, rank, ierr)) 27 28! Create a parallel vector. 29! - In this case, we specify the size of the local portion on 30! each processor, and PETSc computes the global size. Alternatively, 31! if we pass the global size and use PETSC_DECIDE for the 32! local size PETSc will choose a reasonable partition trying 33! to put nearly an equal number of elements on each processor. 34 35 N = rank + 1 36 ione = 1 37 PetscCallA(VecCreateFromOptions(PETSC_COMM_WORLD, PETSC_NULL_CHARACTER, ione, N, PETSC_DECIDE, x, ierr)) 38 PetscCallA(VecGetSize(x, N, ierr)) 39 PetscCallA(VecSet(x, one, ierr)) 40 41! Set the vector elements. 42! - Note that VecSetValues() uses 0-based row and column numbers 43! in Fortran as well as in C. 44! - Always specify global locations of vector entries. 45! - Each processor can contribute any vector entries, 46! regardless of which processor "owns" them; any nonlocal 47! contributions will be transferred to the appropriate processor 48! during the assembly process. 49! - In this example, the flag ADD_VALUES indicates that all 50! contributions will be added together. 51 52 ione = 1 53 do 100 i = 0, N - rank - 1 54 PetscCallA(VecSetValues(x, ione, [i], [one], ADD_VALUES, ierr)) 55100 continue 56 57! Assemble vector, using the 2-step process: 58! VecAssemblyBegin(), VecAssemblyEnd() 59! Computations can be done while messages are in transition 60! by placing code between these two statements. 61 62 PetscCallA(VecAssemblyBegin(x, ierr)) 63 PetscCallA(VecAssemblyEnd(x, ierr)) 64 65! Test VecGetValues() with scalar entries 66 if (rank == 0) then 67 ione = 0 68 PetscCallA(VecGetValues(x, ione, [i], value, ierr)) 69 end if 70 71! View the vector; then destroy it. 72 73 PetscCallA(VecView(x, PETSC_VIEWER_STDOUT_WORLD, ierr)) 74 PetscCallA(VecDestroy(x, ierr)) 75 76 PetscCallA(PetscFinalize(ierr)) 77 end 78 79!/*TEST 80! 81! test: 82! nsize: 2 83! filter: grep -v " MPI process" 84! 85!TEST*/ 86