Fortran 阻止执行，直到通过MPI_Comm_spawn调用的子项完成，否则无法访问子项

我想从main.f90代码生成child.exe，但我没有修改child.f90的权限。简化的代码如下所示：

main.f90:

program main
   implicit none
   include 'mpif.h'
   integer, parameter:: root = 0
   integer, dimension(:), allocatable:: errCodes, status
   integer:: numprocs, request, ierr, taskid, INTERCOMM
   character*(*), PARAMETER:: cmd = "./child.exe"
!
   call mpi_init(ierr)
   call mpi_comm_rank(mpi_comm_world, taskid, ierr)
   call mpi_comm_size(mpi_comm_world, numprocs, ierr )
   call mpi_barrier(mpi_comm_world, ierr)
!   
   allocate( errCodes(numprocs), status(numprocs))
   print*, "from parent, CPU: ", taskid
   call MPI_COMM_SPAWN(cmd, MPI_ARGV_NULL, 2, MPI_INFO_NULL, root, MPI_COMM_WORLD,INTERCOMM, errCodes, ierr)
!
   call mpi_barrier(mpi_comm_world, ierr)
!
   print*, "After children, from parent CPU", taskid
   call mpi_finalize(ierr)
end program main

 program child
   implicit none
   include 'mpif.h'
   integer:: numprocs, temp, ierr, arg_count, taskid, maxproces, errCodes(2)
!  
   call mpi_init(ierr)
   call mpi_comm_rank(mpi_comm_world, taskid, ierr)
   call mpi_comm_size(mpi_comm_world, numprocs, ierr )
!   
   call sleep(1)
!
   print*, "from child, CPU: ", taskid
   call mpi_finalize(ierr)
end program child

child.f90:

program main
   implicit none
   include 'mpif.h'
   integer, parameter:: root = 0
   integer, dimension(:), allocatable:: errCodes, status
   integer:: numprocs, request, ierr, taskid, INTERCOMM
   character*(*), PARAMETER:: cmd = "./child.exe"
!
   call mpi_init(ierr)
   call mpi_comm_rank(mpi_comm_world, taskid, ierr)
   call mpi_comm_size(mpi_comm_world, numprocs, ierr )
   call mpi_barrier(mpi_comm_world, ierr)
!   
   allocate( errCodes(numprocs), status(numprocs))
   print*, "from parent, CPU: ", taskid
   call MPI_COMM_SPAWN(cmd, MPI_ARGV_NULL, 2, MPI_INFO_NULL, root, MPI_COMM_WORLD,INTERCOMM, errCodes, ierr)
!
   call mpi_barrier(mpi_comm_world, ierr)
!
   print*, "After children, from parent CPU", taskid
   call mpi_finalize(ierr)
end program main

 program child
   implicit none
   include 'mpif.h'
   integer:: numprocs, temp, ierr, arg_count, taskid, maxproces, errCodes(2)
!  
   call mpi_init(ierr)
   call mpi_comm_rank(mpi_comm_world, taskid, ierr)
   call mpi_comm_size(mpi_comm_world, numprocs, ierr )
!   
   call sleep(1)
!
   print*, "from child, CPU: ", taskid
   call mpi_finalize(ierr)
end program child

---

我的问题与类似，但如果有人能澄清，如果我无法接触到孩子，该如何处理，我将不胜感激。我曾尝试使用mpi_屏障，但没有任何帮助。

对我来说，Hristo从链接答案中得到的建议效果很好，只需密切遵循即可。注意通讯器（

intercom

，

parent

）。最后仍然存在一些问题，我以前从未使用过子进程，但它应该为您指出正确的方向

program main
   use mpi
   implicit none
   integer, parameter:: root = 0
   integer, dimension(:), allocatable:: errCodes, status
   integer:: numprocs, request, ierr, taskid, INTERCOMM
   character*(*), PARAMETER:: cmd = "./child.exe"
!
   call mpi_init(ierr)
   call mpi_comm_rank(mpi_comm_world, taskid, ierr)
   call mpi_comm_size(mpi_comm_world, numprocs, ierr )
   call mpi_barrier(mpi_comm_world, ierr)
!   
   allocate( errCodes(numprocs), status(numprocs))
   print*, "from parent, CPU: ", taskid
   call MPI_COMM_SPAWN(cmd, MPI_ARGV_NULL, 1, MPI_INFO_NULL, root, MPI_COMM_WORLD, INTERCOMM, errCodes, ierr)
!
   call mpi_barrier(INTERCOMM, ierr)
!
   call mpi_barrier(INTERCOMM, ierr)

   print*, "After children, from parent CPU", taskid

   call mpi_finalize(ierr)
end program main


 program child
   use mpi
   implicit none
   integer:: numprocs, temp, ierr, arg_count, taskid, maxproces, errCodes(2), parent
!  
   call mpi_init(ierr)
   call mpi_Comm_get_parent(parent, ierr)
   call mpi_comm_rank(parent, taskid, ierr)
   call mpi_comm_size(parent, numprocs, ierr )
!   
   call sleep(5)
   call mpi_barrier(parent, ierr)
!
   print*, "from child, CPU: ", taskid
   call sleep(1)

   call mpi_barrier(parent, ierr)

!    call sleep(2)
   call mpi_finalize(ierr)
end program child

---

什么意思：没有帮助“？顺便说一句，include

mpif.h

更多的是为了与FORTRAN 77兼容，我建议在FORTRAN 90及更新版本中使用

mpi

。我添加了一个mpi_屏障，以便我可以在子级打印消息之后从主级查看打印消息。实际上，主程序并没有等待孩子完成它的执行。看到答案了吗，你的通讯器错了。我猜它有问题。当我使用“mpirun-np 2 a.out”运行它时，它会在调用子项之前继续打印main.f90中的第一条消息，而不会在生成后打印child.f90中的消息或main.f90中的消息。为什么要使用

-np 2

运行它？它实际上对我来说甚至适用于

-np 2

，但我不能说它在任何地方都适用。当然，它会打印第一行两次。为了能够设置我想要使用的CPU数量。即使我使用.out来运行main.f90生成的可执行文件，屏幕上也会显示大量“来自父级，CPU:0”。这是一个测试代码，使用多个处理器调用它可能没有意义。