MPI 代码在 MPI_Comm_split 之后返回无效的子通信器

Question

我在使用 Fortran，我正在学习 MPI。具体来说，我正在使用 4 个进程，我想通过创建一个子通信器只向等级 0、等级 0、等级 0、等级 1、等级 2 发送消息，以练习它们的使用。无论我尝试什么，我都没有得到共享消息的预期结果。

这是我试过的。

  SUBROUTINE FOO(IER)

  USE MPI
  
  INTEGER    :: IER

  integer :: ierr, rank, size, color
  integer :: subcomm
  integer :: msg

  call MPI_Init(ierr)
  call MPI_Comm_rank(MPI_COMM_WORLD, rank, ierr)
  call MPI_Comm_size(MPI_COMM_WORLD, size, ierr)
  
  color = MPI_UNDEFINED
  msg = -1
  ! Split the original communicator into two subcommunicators based on rank
  if (rank < 3) then
     color = 0 ! First subcomm has color 0
  end if
  if (rank == 0) then
     msg = 100
  endif

  ! Create the subcommunicator
  call MPI_Comm_split(MPI_COMM_WORLD, color, rank, subcomm, ierr)
  
  if (rank < 3) then 
     ! Broadcast the message within each subcomm
     call MPI_Bcast(msg, 1, MPI_INTEGER, 0, subcomm, ierr)

     ! Print the result on each process
     print *, 'Process ', rank, ' of ', size, ' in subcomm received msg = ', msg
  endif

  ! Free the subcommunicator and finalize MPI
  call MPI_Comm_free(subcomm, ierr)
  call MPI_Finalize(ierr)

  END SUBROUTINE

函数FOO被主程序调用：

  PROGRAM DUMMY

  IMPLICIT NONE
  
  INTEGER          IER
  
  CALL FOO(IER)
  
  END PROGRAM DUMMY

错误信息发生在运行时：

Process 0  of 4  in subcomm received msg = 100
Process 2 of 4  in subcomm received msg = -1
Process 1  of 4  in subcomm received msg = -1

subcomm 顺便说一句，总是等于 0。

Answer 1

这条线：

call MPI_Comm_free(subcomm, ierr)

在 subcommunicator 为空的进程上给出错误。将相同的条件放在该行上，就像您放在 split call 和 bcast 上一样。下次请附上错误信息。