MPI_COMM_SPAWN_MULTIPLE(3)

πŸ’– NAME πŸ’–

MPI_Comm_spawn_multiple β€” okay so picture MPI_Comm_spawn but it can do a whole squad: it spawns multiple binaries, or the same binary with multiple sets of arguments. It's giving main character energy for process launching. ✨

This is the Open MPI page, bestie, so you're in the right place. 🫢


✨ SYNTAX ✨

C Syntax

#include <mpi.h>

int MPI_Comm_spawn_multiple(int count, char *array_of_commands[],
     char **array_of_argv[], const int array_of_maxprocs[], const MPI_Info
     array_of_info[], int root, MPI_Comm comm, MPI_Comm *intercomm,
     int array_of_errcodes[])

Fortran Syntax

USE MPI
! or the older form: INCLUDE 'mpif.h'
MPI_COMM_SPAWN_MULTIPLE(COUNT, ARRAY_OF_COMMANDS, ARRAY_OF_ARGV,
     ARRAY_OF_MAXPROCS, ARRAY_OF_INFO, ROOT, COMM, INTERCOMM,
     ARRAY_OF_ERRCODES, IERROR)
     INTEGER COUNT, ARRAY_OF_INFO(*), ARRAY_OF_MAXPROCS(*), ROOT,
             COMM, INTERCOMM, ARRAY_OF_ERRCODES(*), IERROR
     CHARACTER*(*) ARRAY_OF_COMMANDS(*), ARRAY_OF_ARGV(COUNT, *)

Fortran 2008 Syntax

USE mpi_f08
MPI_Comm_spawn_multiple(count, array_of_commands, array_of_argv,
     array_of_maxprocs, array_of_info, root, comm, intercomm,
             array_of_errcodes, ierror)
     INTEGER, INTENT(IN) :: count, array_of_maxprocs(*), root
     CHARACTER(LEN=*), INTENT(IN) :: array_of_commands(*)
     CHARACTER(LEN=*), INTENT(IN) :: array_of_argv(count, *)
     TYPE(MPI_Info), INTENT(IN) :: array_of_info(*)
     TYPE(MPI_Comm), INTENT(IN) :: comm
     TYPE(MPI_Comm), INTENT(OUT) :: intercomm
     INTEGER :: array_of_errcodes(*)
     INTEGER, OPTIONAL, INTENT(OUT) :: ierror

🌸 INPUT PARAMETERS 🌸

  • count β€” the number of commands. A positive integer, and it's only significant to MPI over at root β€” peep the NOTES on that one. πŸ‘€
  • array_of_commands β€” the programs to be executed. Array of strings, and again, only significant at root.
  • array_of_argv β€” the arguments for those commands. Array of array of strings, significant only at root.
  • array_of_maxprocs β€” the maximum number of processes to start for each command. Array of integers, significant only at root.
  • array_of_info β€” info objects telling the runtime system where and how to start processes. Array of handles, significant only at root.
  • root β€” the rank of the process where all those previous arguments actually get examined. Integer vibes.
  • comm β€” the intracommunicator holding the group of spawning processes. A handle.

πŸ¦‹ OUTPUT PARAMETERS πŸ¦‹

  • intercomm β€” the intercommunicator between the original group and the newly spawned group. A handle, bestie, this is your connection to the babies. πŸ’«
  • array_of_errcodes β€” one code per process. Array of integers.
  • ierror β€” Fortran only: the error status. Integer.

πŸ‘‘ DESCRIPTION πŸ‘‘

MPI_Comm_spawn_multiple is literally identical to MPI_Comm_spawn except that it can specify multiple executables. The first argument, count, tells you how many executables we're working with. The next three arguments are arrays of the corresponding arguments in MPI_Comm_spawn. Then array_of_info is an array of info arguments, one for each executable β€” go read the INFO ARGUMENTS section for the tea on that. 🍡

For the Fortran version of array_of_argv, the element array_of_argv(i,j) is the jth argument to command number i. Simple as.

In any language, an application may use the constant MPI_ARGVS_NULL (which is likely to be (char **)0 in C) to specify that no arguments should be passed to any commands. Now listen β€” the effect of setting individual elements of array_of_argv to MPI_ARGV_NULL is not defined. Undefined behaviour, babes, do not do it, I am begging. 😭 To specify arguments for some commands but not others, the commands without arguments should have a corresponding argv whose first element is null ((char *)0 in C and empty string in Fortran).

All of the spawned processes have the same MPI_COMM_WORLD. Their ranks in MPI_COMM_WORLD correspond directly to the order in which the commands are specified in MPI_Comm_spawn_multiple. So assume that m1 processes are generated by the first command, m2 by the second, etc.:

  • The processes corresponding to the first command have ranks 0, 1,…, m1-1.
  • The processes in the second command have ranks m1, m1+1, …, m1+m2-1.
  • The processes in the third have ranks m1+m2, m1+m2+1, …, m1+m2+m3-1, etc.

The array_of_errcodes argument is a 1-dimensional array of size

     _ count
    \       n ,
    /_ i=1   i

where i is the ith element of array_of_maxprocs. Command number i corresponds to the i contiguous slots in this array from element

                 _              _
 _ i-1          |   _ i          |
\       n ,  to |  \      n      | -1
/_ j=1   i      |  /_ j=1  j     |
                |_              _|

Error codes are treated as for MPI_Comm_spawn. πŸ’…


πŸŽ€ INFO ARGUMENTS πŸŽ€

These are the keys for info that are recognized in #PACKAGE_NAME#. Just so you know: the reserved values mentioned in Section 5.3.4 of the MPI-2 standard are not implemented. 🚫

  • host (char *) β€” Comma-separated list of hosts on which the processes should be spawned. See the orte_host man page for an explanation of how this will be used.
  • hostfile (char *) β€” Hostfile containing the hosts on which the processes are to be spawned. See the orte_hostfile man page for an explanation of how this will be used.
  • add-host (char *) β€” Add the specified hosts to the list of hosts known to this job and use it for the associated processes. This will be used similarly to the -host option.
  • add-hostfile (char *) β€” Hostfile containing hosts to be added to the list of hosts known to this job and use it for the associated process. This will be used similarly to the -hostfile option.
  • wdir (char *) β€” Directory where the executable is located. If files are to be pre-positioned, then this location is the desired working directory at time of execution - if not specified, then it will automatically be set to ompi_preload_files_dest_dir.
  • ompi_prefix (char *) β€” Same as the --prefix command line argument to mpirun.
  • ompi_preload_binary (bool) β€” If set to true, pre-position the specified executable onto the remote host. A destination directory must also be provided.
  • ompi_preload_files (char *) β€” A comma-separated list of files that are to be pre-positioned in addition to the executable. Note that this option does not depend upon ompi_preload_binary - files can be moved to the target even if an executable is not moved.
  • ompi_stdin_target (char *) β€” Comma-delimited list of ranks to receive stdin when forwarded.
  • ompi_non_mpi (bool) β€” If set to true, launching a non-MPI application; the returned communicator will be MPI_COMM_NULL. Failure to set this flag when launching a non-MPI application will cause both the child and parent jobs to "hang". ⚠️
  • ompi_param (char *) β€” Pass an OMPI MCA parameter to the child job. If that parameter already exists in the environment, the value will be overwritten by the provided value.
  • mapper (char *) β€” Mapper to be used for this job
  • map_by (char *) β€” Mapping directive indicating how processes are to be mapped (slot, node, socket, etc.).
  • rank_by (char *) β€” Ranking directive indicating how processes are to be ranked (slot, node, socket, etc.).
  • bind_to (char *) β€” Binding directive indicating how processes are to be bound (core, slot, node, socket, etc.).
  • path (char *) β€” List of directories to search for the executable
  • npernode (char *) β€” Number of processes to spawn on each node of the allocation
  • pernode (bool) β€” Equivalent to npernode of 1
  • ppr (char *) β€” Spawn specified number of processes on each of the identified object type
  • env (char *) β€” Newline-delimited list of envars to be passed to the spawned procs

bool info keys are actually strings but are evaluated as follows, bestie, pay attention: if the string value is a number, it is converted to an integer and cast to a boolean (meaning that zero integers are false and non-zero values are true). If the string value is (case-insensitive) β€œyes” or β€œtrue”, the boolean is true. If the string value is (case-insensitive) β€œno” or β€œfalse”, the boolean is false. All other string values are unrecognized, and therefore false. That's it, that's the whole vibe check. βœ…βŒ

Note that if any of the info handles have ompi_non_mpi set to true, then all info handles must have it set to true. If some are set to true, but others are set to false (or are unset), MPI_ERR_INFO will be returned. No mixing, no matching. 🚨

Note that in #PACKAGE_NAME#, the first array location in array_of_info is applied to all the commands in array_of_commands. πŸ‘‘


πŸ’… NOTES πŸ’…

okay so bestie, listen up because this one has ✨lore✨.

The argument count is interpreted by MPI only at the root, and same goes for array_of_argv. And here's the lil gotcha: since the leading dimension of array_of_argv is count, a nonpositive value of count at a nonroot node could theoretically cause a runtime bounds check error β€” even though array_of_argv should be totally ignored by the subroutine. πŸ™ˆ so if that happens to you, just explicitly supply a reasonable value of count on the nonroot nodes. problem solved, no drama. πŸ’…

Similar to MPI_Comm_spawn, it is the application's responsibility to terminate each individual set of argv in the array_of_argv argument. so like, you gotta do it yourself, no one's coming to save you:

  • In C, each argv array is terminated by a NULL pointer.
  • In Fortran, each argv array is terminated by an empty string β€” and bestie, note that compilers will not automatically insert this blank string; the application must ensure to have enough space for an empty string entry as the last element of the array. 😭

Other restrictions apply to the array_of_argv parameter; see MPI_Comm_spawn's description of the argv parameter for more details. iykyk. 🫢

MPI-3.1 implies (but does not directly state) that the argument array_of_commands must be an array of strings of length count. Unlike the array_of_argv parameter, array_of_commands does not need to be terminated with a NULL pointer in C or a blank string in Fortran. πŸŽ€ Older versions of Open MPI did require that array_of_commands be terminated with a blank string in Fortran β€” that is no longer required in this version of Open MPI. glow up fr fr. ✨

And here's the tea β˜• β€” calling MPI_Comm_spawn many times would create many sets of children with different MPI_COMM_WORLDs, whereas MPI_Comm_spawn_multiple creates children with a single MPI_COMM_WORLD, so the two methods are not completely equivalent. Also, if you need to spawn multiple executables, you may get better performance by using MPI_Comm_spawn_multiple instead of calling MPI_Comm_spawn several times. efficiency queen behavior, honestly. πŸ‘‘


πŸ”₯ ERRORS πŸ”₯

### ⚠️ Important In Open MPI, this function only works properly when the MPI job is launched via mpirun / mpiexec. Using this function in other run-time environments is currently unsupported, and may result in undefined behavior (e.g., the MPI job may hang). πŸ˜­πŸ’€ so like. don't do that. fr.

Almost all MPI routines return an error value; C routines as the return result of the function and Fortran routines in the last argument. classic duality. πŸ’…

Before the error value is returned, the current MPI error handler associated with the communication object (e.g., communicator, window, file) is called. If no communication object is associated with the MPI call, then the call is considered attached to MPI_COMM_SELF and will call the associated MPI error handler. When MPI_COMM_SELF is not initialized (i.e., before MPI_Init/MPI_Init_thread, after MPI_Finalize, or when using the Sessions Model exclusively) the error raises the initial error handler. The initial error handler can be changed by calling MPI_Comm_set_errhandler on MPI_COMM_SELF when using the World model, or the mpi_initial_errhandler CLI argument to mpiexec or info key to MPI_Comm_spawn/MPI_Comm_spawn_multiple. If no other appropriate error handler has been set, then the MPI_ERRORS_RETURN error handler is called for MPI I/O functions and the MPI_ERRORS_ABORT error handler is called for all other MPI functions. okay that was a lot but we got through it together bestie. 🫢

Open MPI includes three predefined error handlers that can be used:

  • MPI_ERRORS_ARE_FATAL β€” Causes the program to abort all connected MPI processes. πŸ’€
  • MPI_ERRORS_ABORT β€” An error handler that can be invoked on a communicator, window, file, or session. When called on a communicator, it acts as if MPI_Abort was called on that communicator. If called on a window or file, acts as if MPI_Abort was called on a communicator containing the group of processes in the corresponding window or file. If called on a session, aborts only the local process. ✨
  • MPI_ERRORS_RETURN β€” Returns an error code to the application. the chill one. 😌

MPI applications can also implement their own error handlers by calling:

  • MPI_Comm_create_errhandler then MPI_Comm_set_errhandler
  • MPI_File_create_errhandler then MPI_File_set_errhandler
  • MPI_Session_create_errhandler then MPI_Session_set_errhandler or at MPI_Session_init
  • MPI_Win_create_errhandler then MPI_Win_set_errhandler

Note that MPI does not guarantee that an MPI program can continue past an error. bestie. that's the ironclad rule. πŸ™ˆ

See the MPI man page for a full list of MPI error codes.

See the Error Handling section of the MPI-3.1 standard for more information.

Also see:

  • MPI_Comm_spawn
  • MPI_Comm_get_parent
  • mpirun(1)

2003-2026, The Open MPI Community


k that's the whole page bestie, you ate it up honestly πŸ’– the source/version line down at the bottom says MPI_COMM_SPAWN_MULTIPLE(3) and the date is Sep 10, 2026, so go spawn those children (the MPI kind, obviously πŸ¦‹) and slay. periodt. πŸ‘‘βœ¨

πŸ“¦ where this page comes from