2004-01-10 01:09:51 +03:00
|
|
|
/*
|
2005-11-05 22:57:48 +03:00
|
|
|
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
|
|
|
|
* University Research and Technology
|
|
|
|
* Corporation. All rights reserved.
|
|
|
|
* Copyright (c) 2004-2005 The University of Tennessee and The University
|
|
|
|
* of Tennessee Research Foundation. All rights
|
|
|
|
* reserved.
|
2004-11-28 23:09:25 +03:00
|
|
|
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
|
|
|
* University of Stuttgart. All rights reserved.
|
2005-03-24 15:43:37 +03:00
|
|
|
* Copyright (c) 2004-2005 The Regents of the University of California.
|
|
|
|
* All rights reserved.
|
2006-05-11 23:46:21 +04:00
|
|
|
* Copyright (c) 2006 Cisco Systems, Inc. All rights reserved.
|
2007-07-26 01:01:10 +04:00
|
|
|
* Copyright (c) 2007 Los Alamos National Security, LLC. All rights
|
|
|
|
* reserved.
|
2004-11-22 04:38:40 +03:00
|
|
|
* $COPYRIGHT$
|
|
|
|
*
|
|
|
|
* Additional copyrights may follow
|
|
|
|
*
|
2004-01-10 01:09:51 +03:00
|
|
|
* $HEADER$
|
|
|
|
*/
|
|
|
|
|
2007-07-26 01:01:10 +04:00
|
|
|
|
|
|
|
/** @file
|
|
|
|
* Process identification structure interface
|
|
|
|
*
|
|
|
|
* Process identification structure interface. The ompi_proc_t
|
|
|
|
* structure contatins basic information about the remote (and local)
|
|
|
|
* processes.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#ifndef OMPI_PROC_PROC_H
|
|
|
|
#define OMPI_PROC_PROC_H
|
2004-01-10 01:09:51 +03:00
|
|
|
|
2006-02-12 04:33:29 +03:00
|
|
|
#include "ompi/types.h"
|
2005-07-03 20:22:16 +04:00
|
|
|
#include "opal/class/opal_list.h"
|
2006-02-12 04:33:29 +03:00
|
|
|
#include "orte/dss/dss_types.h"
|
2005-07-04 02:45:48 +04:00
|
|
|
#include "opal/threads/mutex.h"
|
2005-03-14 23:57:21 +03:00
|
|
|
|
2006-02-12 04:33:29 +03:00
|
|
|
#include "orte/mca/ns/ns_types.h"
|
2005-03-14 23:57:21 +03:00
|
|
|
|
2007-07-26 01:01:10 +04:00
|
|
|
BEGIN_C_DECLS
|
2004-01-10 01:09:51 +03:00
|
|
|
|
2007-07-26 01:01:10 +04:00
|
|
|
/* ******************************************************************** */
|
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Remote Open MPI process structure
|
|
|
|
*
|
|
|
|
* Remote Open MPI process structure. Each process contains exactly
|
|
|
|
* one ompi_proc_t structure for each remote process it knows about.
|
|
|
|
*/
|
2004-06-07 19:33:53 +04:00
|
|
|
struct ompi_proc_t {
|
2005-07-15 02:43:01 +04:00
|
|
|
/** allow proc to be placed on a list */
|
2007-07-26 01:01:10 +04:00
|
|
|
opal_list_item_t super;
|
2005-07-15 02:43:01 +04:00
|
|
|
/** this process' name */
|
2007-07-26 01:01:10 +04:00
|
|
|
orte_process_name_t proc_name;
|
2005-07-15 02:43:01 +04:00
|
|
|
/** PML specific proc data */
|
2006-07-04 05:20:20 +04:00
|
|
|
struct mca_pml_base_endpoint_t* proc_pml;
|
2007-07-26 01:01:10 +04:00
|
|
|
/** BML specific proc data */
|
|
|
|
struct mca_bml_base_endpoint_t* proc_bml;
|
2005-07-15 02:43:01 +04:00
|
|
|
/** MCA module exchange data */
|
2007-07-26 01:01:10 +04:00
|
|
|
opal_object_t* proc_modex;
|
2005-07-15 02:43:01 +04:00
|
|
|
/** architecture of this process */
|
2007-07-26 01:01:10 +04:00
|
|
|
uint32_t proc_arch;
|
|
|
|
/** Base convertor for the proc described by this process */
|
|
|
|
struct ompi_convertor_t* proc_convertor;
|
|
|
|
/** Lock protecting data inside the given ompi_proc_t */
|
|
|
|
opal_mutex_t proc_lock;
|
2006-05-11 23:46:21 +04:00
|
|
|
/** Keep the hostname around for debugging purposes */
|
2007-07-26 01:01:10 +04:00
|
|
|
char* proc_hostname;
|
2005-07-15 02:43:01 +04:00
|
|
|
/** flags for this proc */
|
2007-07-26 01:01:10 +04:00
|
|
|
uint8_t proc_flags;
|
2004-01-10 01:09:51 +03:00
|
|
|
};
|
2004-06-07 19:33:53 +04:00
|
|
|
typedef struct ompi_proc_t ompi_proc_t;
|
2007-07-26 01:01:10 +04:00
|
|
|
OBJ_CLASS_DECLARATION(ompi_proc_t);
|
2004-01-10 01:09:51 +03:00
|
|
|
|
2005-07-15 02:43:01 +04:00
|
|
|
|
|
|
|
/**
|
2007-07-26 01:01:10 +04:00
|
|
|
* @private
|
|
|
|
*
|
|
|
|
* Pointer to the ompi_proc_t structure for the local process
|
|
|
|
*
|
|
|
|
* Pointer to the ompi_proc_t structure for the local process.
|
|
|
|
*
|
|
|
|
* @note This pointer is declared here to allow inline functions
|
|
|
|
* within this header file to access the local process quickly.
|
|
|
|
* Please use ompi_proc_local() instead.
|
2005-07-15 02:43:01 +04:00
|
|
|
*/
|
2007-07-26 01:01:10 +04:00
|
|
|
OMPI_DECLSPEC extern ompi_proc_t* ompi_proc_local_proc;
|
|
|
|
|
|
|
|
|
|
|
|
/* ******************************************************************** */
|
|
|
|
|
|
|
|
|
|
|
|
/** Process is on the same node as the local process */
|
2005-07-15 02:43:01 +04:00
|
|
|
#define OMPI_PROC_FLAG_LOCAL 0x01
|
|
|
|
|
2004-01-10 01:09:51 +03:00
|
|
|
|
2007-07-26 01:01:10 +04:00
|
|
|
/* ******************************************************************** */
|
|
|
|
|
|
|
|
|
2004-02-13 16:56:55 +03:00
|
|
|
/**
|
2007-07-26 01:01:10 +04:00
|
|
|
* Initialize the OMPI process subsystem
|
|
|
|
*
|
|
|
|
* Initialize the Open MPI process subsystem. This function will
|
|
|
|
* query the run-time environment and build a list of the proc
|
|
|
|
* instances in the current MPI_COMM_WORLD. The local information not
|
|
|
|
* easily determined by the run-time ahead of time (architecture and
|
|
|
|
* hostname) will be published during this call.
|
|
|
|
*
|
|
|
|
* @note While an ompi_proc_t will exist with mostly valid information
|
|
|
|
* for each process in the MPI_COMM_WORLD at the conclusion of this
|
|
|
|
* call, some information will not be immediately available. This
|
|
|
|
* includes the architecture and hostname, which will be available by
|
|
|
|
* the conclusion of the stage gate.
|
|
|
|
*
|
|
|
|
* @retval OMPI_SUCESS System successfully initialized
|
|
|
|
* @retval OMPI_ERRROR Initialization failed due to unspecified error
|
2004-02-13 16:56:55 +03:00
|
|
|
*/
|
2004-06-07 19:33:53 +04:00
|
|
|
int ompi_proc_init(void);
|
2004-02-13 16:56:55 +03:00
|
|
|
|
2007-07-26 01:01:10 +04:00
|
|
|
|
2007-08-09 22:53:28 +04:00
|
|
|
/**
|
|
|
|
* Get data exchange information from remote processes
|
|
|
|
*
|
|
|
|
* Get data exchanged from remote processes and populate the ompi proc
|
|
|
|
* structures for the associated processes.
|
|
|
|
*
|
|
|
|
* @retval OMPI_SUCCESS Information successfully received
|
|
|
|
* @retval OMPI_ERROR Information update failure
|
|
|
|
*/
|
|
|
|
int ompi_proc_get_info(void);
|
|
|
|
|
|
|
|
|
2004-12-02 16:28:10 +03:00
|
|
|
/**
|
2007-07-26 01:01:10 +04:00
|
|
|
* Finalize the OMPI Process subsystem
|
|
|
|
*
|
|
|
|
* Finalize the Open MPI process subsystem. This function will
|
|
|
|
* release all memory created during the life of the application,
|
|
|
|
* including all ompi_proc_t structures.
|
|
|
|
*
|
|
|
|
* @retval OMPI_SUCCESS System successfully finalized
|
2004-12-02 16:28:10 +03:00
|
|
|
*/
|
|
|
|
int ompi_proc_finalize(void);
|
|
|
|
|
2007-07-26 01:01:10 +04:00
|
|
|
|
2004-02-13 16:56:55 +03:00
|
|
|
/**
|
|
|
|
* Returns the list of proc instances associated with this job.
|
2007-07-26 01:01:10 +04:00
|
|
|
*
|
|
|
|
* Returns the list of proc instances associated with this job. Given
|
|
|
|
* the current association between a job and an MPI_COMM_WORLD, this
|
|
|
|
* function provides the process instances for the current
|
|
|
|
* MPI_COMM_WORLD.
|
|
|
|
*
|
|
|
|
* @note The reference count of each process in the array is
|
|
|
|
* incremented and the caller is responsible for releasing each
|
|
|
|
* process in the array, as well as freeing the array.
|
|
|
|
*
|
|
|
|
* @param[in] size Number of processes in the ompi_proc_t array
|
|
|
|
*
|
|
|
|
* @return Array of pointers to proc instances in the current
|
|
|
|
* MPI_COMM_WORLD, or NULL if there is an internal failure.
|
2004-02-13 16:56:55 +03:00
|
|
|
*/
|
2007-02-27 18:17:17 +03:00
|
|
|
OMPI_DECLSPEC ompi_proc_t** ompi_proc_world(size_t* size);
|
2004-02-13 16:56:55 +03:00
|
|
|
|
2007-07-26 01:01:10 +04:00
|
|
|
|
2004-02-13 16:56:55 +03:00
|
|
|
/**
|
|
|
|
* Returns the list of all known proc instances.
|
2007-07-26 01:01:10 +04:00
|
|
|
*
|
|
|
|
* Returns the list of all known proc instances, including those in
|
|
|
|
* other MPI_COMM_WORLDs. It is possible that we may no longer be
|
|
|
|
* connected to some of the procs returned (in the MPI sense of the
|
|
|
|
* word connected). In a strictly MPI-1 application, this function
|
|
|
|
* will return the same information as ompi_proc_world().
|
|
|
|
*
|
|
|
|
* @note The reference count of each process in the array is
|
|
|
|
* incremented and the caller is responsible for releasing each
|
|
|
|
* process in the array, as well as freeing the array.
|
|
|
|
*
|
|
|
|
* @param[in] size Number of processes in the ompi_proc_t array
|
|
|
|
*
|
|
|
|
* @return Array of pointers to proc instances in the current
|
|
|
|
* known universe, or NULL if there is an internal failure.
|
2004-02-13 16:56:55 +03:00
|
|
|
*/
|
2007-02-27 18:17:17 +03:00
|
|
|
OMPI_DECLSPEC ompi_proc_t** ompi_proc_all(size_t* size);
|
2004-02-13 16:56:55 +03:00
|
|
|
|
2007-07-26 01:01:10 +04:00
|
|
|
|
2004-02-13 16:56:55 +03:00
|
|
|
/**
|
2007-07-26 01:01:10 +04:00
|
|
|
* Returns a list of the local process
|
|
|
|
*
|
|
|
|
* Returns a list containing the local process (and only the local
|
|
|
|
* process). Has calling semantics similar to ompi_proc_world() and
|
|
|
|
* ompi_proc_all().
|
|
|
|
*
|
|
|
|
* @note The reference count of each process in the array is
|
|
|
|
* incremented and the caller is responsible for releasing each
|
|
|
|
* process in the array, as well as freeing the array.
|
|
|
|
*
|
|
|
|
* @param[in] size Number of processes in the ompi_proc_t array
|
|
|
|
*
|
|
|
|
* @return Array of pointers to proc instances in the current
|
|
|
|
* known universe, or NULL if there is an internal failure.
|
2004-02-13 16:56:55 +03:00
|
|
|
*/
|
2007-07-26 01:01:10 +04:00
|
|
|
OMPI_DECLSPEC ompi_proc_t** ompi_proc_self(size_t* size);
|
|
|
|
|
2004-01-29 18:34:47 +03:00
|
|
|
|
2004-02-13 16:56:55 +03:00
|
|
|
/**
|
2007-07-26 01:01:10 +04:00
|
|
|
* Returns a pointer to the local process
|
|
|
|
*
|
|
|
|
* Returns a pointer to the local process. Unlike ompi_proc_self(),
|
|
|
|
* the reference count on the local proc instance is not modified by
|
|
|
|
* this function.
|
|
|
|
*
|
|
|
|
* @return Pointer to the local process structure
|
2004-02-13 16:56:55 +03:00
|
|
|
*/
|
2004-06-07 19:33:53 +04:00
|
|
|
static inline ompi_proc_t* ompi_proc_local(void)
|
2004-10-28 22:13:43 +04:00
|
|
|
{
|
2004-06-07 19:33:53 +04:00
|
|
|
return ompi_proc_local_proc;
|
2004-01-29 18:34:47 +03:00
|
|
|
}
|
2004-01-10 01:09:51 +03:00
|
|
|
|
2007-07-26 01:01:10 +04:00
|
|
|
|
2004-05-18 01:28:32 +04:00
|
|
|
/**
|
2004-07-01 18:49:54 +04:00
|
|
|
* Returns the proc instance for a given name
|
2007-07-26 01:01:10 +04:00
|
|
|
*
|
|
|
|
* Returns the proc instance for the specified process name. The
|
|
|
|
* reference count for the proc instance is not incremented by this
|
|
|
|
* function.
|
|
|
|
*
|
|
|
|
* @param[in] name The process name to look for
|
|
|
|
*
|
|
|
|
* @return Pointer to the process instance for \c name
|
2004-05-18 01:28:32 +04:00
|
|
|
*/
|
2007-07-26 01:01:10 +04:00
|
|
|
OMPI_DECLSPEC ompi_proc_t * ompi_proc_find ( const orte_process_name_t* name );
|
2004-08-04 21:05:22 +04:00
|
|
|
|
2004-09-17 14:10:24 +04:00
|
|
|
|
2004-08-04 21:05:22 +04:00
|
|
|
/**
|
2007-07-26 01:01:10 +04:00
|
|
|
* Pack proc list into portable buffer
|
|
|
|
*
|
Clean up the way procs are added to the global process list after MPI_INIT:
* Do not add new procs to the global list during modex callback or
when sharing orte names during accept/connect. For modex, we
cache the modex info for later, in case that proc ever does get
added to the global proc list. For accept/connect orte name
exchange between the roots, we only need the orte name, so no
need to add a proc structure anyway. The procs will be added
to the global process list during the proc exchange later in
the wireup process
* Rename proc_get_namebuf and proc_get_proclist to proc_pack
and proc_unpack and extend them to include all information
needed to build that proc struct on a remote node (which
includes ORTE name, architecture, and hostname). Change
unpack to call pml_add_procs for the entire list of new
procs at once, rather than one at a time.
* Remove ompi_proc_find_and_add from the public proc
interface and make it a private function. This function
would add a half-created proc to the global proc list, so
making it harder to call is a good thing.
This means that there's only two ways to add new procs into the global proc list at this time: During MPI_INIT via the call to ompi_proc_init, where my job is added to the list and via ompi_proc_unpack using a buffer from a packed proc list sent to us by someone else. Currently, this is enough to implement MPI semantics. We can extend the interface more if we like, but that may require HNP communication to get the remote proc information and I wanted to avoid that if at all possible.
Refs trac:564
This commit was SVN r12798.
The following Trac tickets were found above:
Ticket 564 --> https://svn.open-mpi.org/trac/ompi/ticket/564
2006-12-07 22:56:54 +03:00
|
|
|
* This function takes a list of ompi_proc_t pointers (e.g. as given
|
|
|
|
* in groups) and returns a orte buffer containing all information
|
|
|
|
* needed to add the proc to a remote list. This includes the ORTE
|
|
|
|
* process name, the architecture, and the hostname. Ordering is
|
|
|
|
* maintained. The buffer is packed to be sent to a remote node with
|
|
|
|
* different architecture (endian or word size). The buffer can be
|
2007-07-26 01:01:10 +04:00
|
|
|
* dss unloaded to be sent using MPI or send using rml_send_packed().
|
|
|
|
*
|
|
|
|
* @param[in] proclist List of process pointers
|
|
|
|
* @param[in] proclistsize Length of the proclist array
|
|
|
|
* @param[in,out] buf An orte_buffer containing the packed names.
|
|
|
|
* The buffer must be constructed but empty when
|
|
|
|
* passed to this function
|
|
|
|
* @retval OMPI_SUCCESS Success
|
|
|
|
* @retval OMPI_ERROR Unspecified error
|
2004-08-04 21:05:22 +04:00
|
|
|
*/
|
Clean up the way procs are added to the global process list after MPI_INIT:
* Do not add new procs to the global list during modex callback or
when sharing orte names during accept/connect. For modex, we
cache the modex info for later, in case that proc ever does get
added to the global proc list. For accept/connect orte name
exchange between the roots, we only need the orte name, so no
need to add a proc structure anyway. The procs will be added
to the global process list during the proc exchange later in
the wireup process
* Rename proc_get_namebuf and proc_get_proclist to proc_pack
and proc_unpack and extend them to include all information
needed to build that proc struct on a remote node (which
includes ORTE name, architecture, and hostname). Change
unpack to call pml_add_procs for the entire list of new
procs at once, rather than one at a time.
* Remove ompi_proc_find_and_add from the public proc
interface and make it a private function. This function
would add a half-created proc to the global proc list, so
making it harder to call is a good thing.
This means that there's only two ways to add new procs into the global proc list at this time: During MPI_INIT via the call to ompi_proc_init, where my job is added to the list and via ompi_proc_unpack using a buffer from a packed proc list sent to us by someone else. Currently, this is enough to implement MPI semantics. We can extend the interface more if we like, but that may require HNP communication to get the remote proc information and I wanted to avoid that if at all possible.
Refs trac:564
This commit was SVN r12798.
The following Trac tickets were found above:
Ticket 564 --> https://svn.open-mpi.org/trac/ompi/ticket/564
2006-12-07 22:56:54 +03:00
|
|
|
int ompi_proc_pack(ompi_proc_t **proclist, int proclistsize,
|
|
|
|
orte_buffer_t *buf);
|
2004-08-04 21:05:22 +04:00
|
|
|
|
|
|
|
|
|
|
|
/**
|
2007-07-26 01:01:10 +04:00
|
|
|
* Unpack a portable buffer of procs
|
2004-08-04 21:05:22 +04:00
|
|
|
*
|
Clean up the way procs are added to the global process list after MPI_INIT:
* Do not add new procs to the global list during modex callback or
when sharing orte names during accept/connect. For modex, we
cache the modex info for later, in case that proc ever does get
added to the global proc list. For accept/connect orte name
exchange between the roots, we only need the orte name, so no
need to add a proc structure anyway. The procs will be added
to the global process list during the proc exchange later in
the wireup process
* Rename proc_get_namebuf and proc_get_proclist to proc_pack
and proc_unpack and extend them to include all information
needed to build that proc struct on a remote node (which
includes ORTE name, architecture, and hostname). Change
unpack to call pml_add_procs for the entire list of new
procs at once, rather than one at a time.
* Remove ompi_proc_find_and_add from the public proc
interface and make it a private function. This function
would add a half-created proc to the global proc list, so
making it harder to call is a good thing.
This means that there's only two ways to add new procs into the global proc list at this time: During MPI_INIT via the call to ompi_proc_init, where my job is added to the list and via ompi_proc_unpack using a buffer from a packed proc list sent to us by someone else. Currently, this is enough to implement MPI semantics. We can extend the interface more if we like, but that may require HNP communication to get the remote proc information and I wanted to avoid that if at all possible.
Refs trac:564
This commit was SVN r12798.
The following Trac tickets were found above:
Ticket 564 --> https://svn.open-mpi.org/trac/ompi/ticket/564
2006-12-07 22:56:54 +03:00
|
|
|
* This function unpacks a packed list of ompi_proc_t structures and
|
|
|
|
* returns the ordered list of proc structures. If the given proc is
|
|
|
|
* already "known", the architecture and hostname information in the
|
|
|
|
* buffer is ignored. If the proc is "new" to this process, it will
|
|
|
|
* be added to the global list of known procs, with information
|
|
|
|
* provided in the buffer. The lookup actions are always entirely
|
|
|
|
* local. The proclist returned is a list of pointers to all procs in
|
|
|
|
* the buffer, whether they were previously known or are new to this
|
|
|
|
* process. PML_ADD_PROCS will be called on the list of new processes
|
|
|
|
* discovered during this operation.
|
2004-08-04 21:05:22 +04:00
|
|
|
*
|
2007-07-26 01:01:10 +04:00
|
|
|
* @param[in] buf orte_buffer containing the packed names
|
|
|
|
* @param[in] proclistsize number of expected proc-pointres
|
|
|
|
* @param[out] proclist list of process pointers
|
|
|
|
*
|
2004-08-04 21:05:22 +04:00
|
|
|
* Return value:
|
|
|
|
* OMPI_SUCCESS on success
|
|
|
|
* OMPI_ERROR else
|
|
|
|
*/
|
Clean up the way procs are added to the global process list after MPI_INIT:
* Do not add new procs to the global list during modex callback or
when sharing orte names during accept/connect. For modex, we
cache the modex info for later, in case that proc ever does get
added to the global proc list. For accept/connect orte name
exchange between the roots, we only need the orte name, so no
need to add a proc structure anyway. The procs will be added
to the global process list during the proc exchange later in
the wireup process
* Rename proc_get_namebuf and proc_get_proclist to proc_pack
and proc_unpack and extend them to include all information
needed to build that proc struct on a remote node (which
includes ORTE name, architecture, and hostname). Change
unpack to call pml_add_procs for the entire list of new
procs at once, rather than one at a time.
* Remove ompi_proc_find_and_add from the public proc
interface and make it a private function. This function
would add a half-created proc to the global proc list, so
making it harder to call is a good thing.
This means that there's only two ways to add new procs into the global proc list at this time: During MPI_INIT via the call to ompi_proc_init, where my job is added to the list and via ompi_proc_unpack using a buffer from a packed proc list sent to us by someone else. Currently, this is enough to implement MPI semantics. We can extend the interface more if we like, but that may require HNP communication to get the remote proc information and I wanted to avoid that if at all possible.
Refs trac:564
This commit was SVN r12798.
The following Trac tickets were found above:
Ticket 564 --> https://svn.open-mpi.org/trac/ompi/ticket/564
2006-12-07 22:56:54 +03:00
|
|
|
int ompi_proc_unpack(orte_buffer_t *buf, int proclistsize, ompi_proc_t ***proclist);
|
2004-08-04 21:05:22 +04:00
|
|
|
|
|
|
|
|
2007-07-26 01:01:10 +04:00
|
|
|
END_C_DECLS
|
2004-01-10 01:09:51 +03:00
|
|
|
|
2007-07-26 01:01:10 +04:00
|
|
|
#endif /* OMPI_PROC_PROC_H */
|