24c811805f
This change contains a non-mandatory modification
of the MPI-RTE interface. Anyone wishing to support
coprocessors such as the Xeon Phi may wish to add
the required definition and underlying support
****************************************************************
Add locality support for coprocessors such as the Intel Xeon Phi.
Detecting that we are on a coprocessor inside of a host node isn't straightforward. There are no good "hooks" provided for programmatically detecting that "we are on a coprocessor running its own OS", and the ORTE daemon just thinks it is on another node. However, in order to properly use the Phi's public interface for MPI transport, it is necessary that the daemon detect that it is colocated with procs on the host.
So we have to split the locality to separately record "on the same host" vs "on the same board". We already have the board-level locality flag, but not quite enough flexibility to handle this use-case. Thus, do the following:
1. add OPAL_PROC_ON_HOST flag to indicate we share a host, but not necessarily the same board
2. modify OPAL_PROC_ON_NODE to indicate we share both a host AND the same board. Note that we have to modify the OPAL_PROC_ON_LOCAL_NODE macro to explicitly check both conditions
3. add support in opal/mca/hwloc/base/hwloc_base_util.c for the host to check for coprocessors, and for daemons to check to see if they are on a coprocessor. The former is done via hwloc, but support for the latter is not yet provided by hwloc. So the code for detecting we are on a coprocessor currently is Xeon Phi specific - hopefully, we will find more generic methods in the future.
4. modify the orted and the hnp startup so they check for coprocessors and to see if they are on a coprocessor, and have the orteds pass that info back in their callback message. Automatically detect that coprocessors have been found and identify which coprocessors are on which hosts. Note that this algo isn't scalable at the moment - this will hopefully be improved over time.
5. modify the ompi proc locality detection function to look for coprocessor host info IF the OMPI_RTE_HOST_ID database key has been defined. RTE's that choose not to provide this support do not have to do anything - the associated code will simply be ignored.
6. include some cleanup of the hwloc open/close code so it conforms to how we did things in other frameworks (e.g., having a single "frame" file instead of open/close). Also, fix the locality flags - e.g., being on the same node means you must also be on the same cluster/cu, so ensure those flags are also set.
cmr:v1.7.4:reviewer=hjelmn
This commit was SVN r29435.
172 строки
7.1 KiB
C
172 строки
7.1 KiB
C
/*
|
|
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
|
|
* University Research and Technology
|
|
* Corporation. All rights reserved.
|
|
* Copyright (c) 2004-2011 The University of Tennessee and The University
|
|
* of Tennessee Research Foundation. All rights
|
|
* reserved.
|
|
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
|
* University of Stuttgart. All rights reserved.
|
|
* Copyright (c) 2004-2005 The Regents of the University of California.
|
|
* All rights reserved.
|
|
* Copyright (c) 2011-2012 Los Alamos National Security, LLC.
|
|
* All rights reserved.
|
|
* Copyright (c) 2013 Intel, Inc. All rights reserved
|
|
* $COPYRIGHT$
|
|
*
|
|
* Additional copyrights may follow
|
|
*
|
|
* $HEADER$
|
|
*/
|
|
|
|
/** @file:
|
|
*
|
|
* Populates global structure with process-specific information.
|
|
*
|
|
*
|
|
*/
|
|
|
|
#ifndef _ORTE_PROC_INFO_H_
|
|
#define _ORTE_PROC_INFO_H_
|
|
|
|
#include "orte_config.h"
|
|
|
|
#ifdef HAVE_STDINT_H
|
|
#include <stdint.h>
|
|
#endif
|
|
|
|
#ifdef HAVE_SYS_TYPES_H
|
|
#include <sys/types.h>
|
|
#endif
|
|
|
|
#include "orte/types.h"
|
|
|
|
#include "opal/dss/dss_types.h"
|
|
#include "opal/mca/hwloc/hwloc.h"
|
|
|
|
#include "orte/mca/grpcomm/grpcomm_types.h"
|
|
|
|
BEGIN_C_DECLS
|
|
|
|
#define ORTE_MAX_HOSTNAME_SIZE 512
|
|
|
|
typedef uint32_t orte_proc_type_t;
|
|
#define ORTE_PROC_TYPE_NONE 0x0000
|
|
#define ORTE_PROC_SINGLETON 0x0001
|
|
#define ORTE_PROC_DAEMON 0x0002
|
|
#define ORTE_PROC_HNP 0x0004
|
|
#define ORTE_PROC_TOOL 0x0008
|
|
#define ORTE_PROC_NON_MPI 0x0010
|
|
#define ORTE_PROC_MPI 0x0020
|
|
#define ORTE_PROC_APP 0x0030
|
|
#define ORTE_PROC_CM 0x0040
|
|
#define ORTE_PROC_CMSLAVE 0x0080
|
|
#define ORTE_PROC_IOF_ENDPT 0x1000
|
|
#define ORTE_PROC_SCHEDULER 0x2000
|
|
|
|
#define ORTE_PROC_IS_SINGLETON (ORTE_PROC_SINGLETON & orte_process_info.proc_type)
|
|
#define ORTE_PROC_IS_DAEMON (ORTE_PROC_DAEMON & orte_process_info.proc_type)
|
|
#define ORTE_PROC_IS_HNP (ORTE_PROC_HNP & orte_process_info.proc_type)
|
|
#define ORTE_PROC_IS_TOOL (ORTE_PROC_TOOL & orte_process_info.proc_type)
|
|
#define ORTE_PROC_IS_NON_MPI (ORTE_PROC_NON_MPI & orte_process_info.proc_type)
|
|
#define ORTE_PROC_IS_MPI (ORTE_PROC_MPI & orte_process_info.proc_type)
|
|
#define ORTE_PROC_IS_APP (ORTE_PROC_APP & orte_process_info.proc_type)
|
|
#define ORTE_PROC_IS_CM (ORTE_PROC_CM & orte_process_info.proc_type)
|
|
#define ORTE_PROC_IS_CMSLAVE (ORTE_PROC_CMSLAVE & orte_process_info.proc_type)
|
|
#define ORTE_PROC_IS_IOF_ENDPT (ORTE_PROC_IOF_ENDPT & orte_process_info.proc_type)
|
|
#define ORTE_PROC_IS_SCHEDULER (ORTE_PROC_SCHEDULER & orte_process_info.proc_type)
|
|
|
|
|
|
/**
|
|
* Process information structure
|
|
*
|
|
* The orte_proc_info() function fills the pid field and obtains the
|
|
* process name, storing that information in the global structure. The
|
|
* structure also holds path names to the universe, job, and process
|
|
* session directories, and to the stdin, stdout, and stderr temp
|
|
* files - however, these are all initialized elsewhere.
|
|
*/
|
|
struct orte_proc_info_t {
|
|
orte_process_name_t my_name; /**< My official process name */
|
|
orte_process_name_t my_daemon; /**< Name of my local daemon */
|
|
char *my_daemon_uri; /**< Contact info to local daemon */
|
|
orte_process_name_t my_hnp; /**< Name of my hnp */
|
|
char *my_hnp_uri; /**< Contact info for my hnp */
|
|
orte_process_name_t my_parent; /**< Name of my parent (or my HNP if no parent was specified) */
|
|
pid_t hnp_pid; /**< hnp pid - used if singleton */
|
|
orte_app_idx_t app_num; /**< our index into the app_context array */
|
|
orte_vpid_t num_procs; /**< number of processes in this job */
|
|
orte_vpid_t max_procs; /**< Maximum number of processes ever in the job */
|
|
orte_vpid_t num_daemons; /**< number of daemons in system */
|
|
int num_nodes; /**< number of nodes in the job */
|
|
char *nodename; /**< string name for this node */
|
|
pid_t pid; /**< Local process ID for this process */
|
|
orte_proc_type_t proc_type; /**< Type of process */
|
|
opal_buffer_t *sync_buf; /**< buffer to store sync response */
|
|
uint16_t my_port; /**< TCP port for out-of-band comm */
|
|
int num_restarts; /**< number of times this proc has restarted */
|
|
orte_node_rank_t my_node_rank; /**< node rank */
|
|
orte_local_rank_t my_local_rank; /**< local rank */
|
|
int32_t num_local_peers; /**< number of procs from my job that share my node with me */
|
|
/* The session directory has the form
|
|
* <prefix>/<openmpi-sessions-user>/<jobid>/<procid>, where the prefix
|
|
* can either be provided by the user via the
|
|
* --tmpdir command-line flag, the use of one of several
|
|
* environmental variables, or else a default location.
|
|
*/
|
|
char *tmpdir_base; /**< Base directory of the session dir tree */
|
|
char *top_session_dir; /**< Top-most directory of the session tree */
|
|
char *job_session_dir; /**< Session directory for job */
|
|
char *proc_session_dir; /**< Session directory for the process */
|
|
|
|
char *sock_stdin; /**< Path name to temp file for stdin. */
|
|
char *sock_stdout; /**< Path name to temp file for stdout. */
|
|
char *sock_stderr; /**< Path name to temp file for stderr. */
|
|
#if OPAL_HAVE_HWLOC
|
|
char *cpuset; /**< String-representation of bitmap where we are bound */
|
|
#endif
|
|
int app_rank; /**< rank within my app_context */
|
|
orte_grpcomm_coll_id_t peer_modex; /**< modex collective id */
|
|
orte_grpcomm_coll_id_t peer_init_barrier; /**< barrier id during init */
|
|
orte_grpcomm_coll_id_t peer_fini_barrier; /**< barrier id during finalize */
|
|
orte_vpid_t my_hostid; /** identifies the local host for a coprocessor */
|
|
};
|
|
typedef struct orte_proc_info_t orte_proc_info_t;
|
|
|
|
|
|
/**
|
|
*
|
|
* Global process info descriptor. Initialized to almost no
|
|
* meaningful information - data is provided by calling \c
|
|
* orte_rte_init() (which calls \c orte_proc_info() to fill in the
|
|
* structure).
|
|
*
|
|
* The exception to this rule is the \c orte_process_info.seed field,
|
|
* which will be initialized to \c false, but should be set to \c true
|
|
* before calling \c orte_rte_info() if the caller is a seed daemon.
|
|
*/
|
|
ORTE_DECLSPEC extern orte_proc_info_t orte_process_info;
|
|
|
|
|
|
/**
|
|
* \internal
|
|
*
|
|
* Global structure to store a wide range of information about the
|
|
* process. orte_proc_info populates a global variable with
|
|
* information about the process being executing. This function should
|
|
* be called only once, from orte_rte_init().
|
|
*
|
|
* @param None.
|
|
*
|
|
* @retval ORTE_SUCCESS Successfully initialized the various fields.
|
|
* @retval OMPI_ERROR Failed to initialize one or more fields.
|
|
*/
|
|
|
|
ORTE_DECLSPEC int orte_proc_info(void);
|
|
|
|
ORTE_DECLSPEC int orte_proc_info_finalize(void);
|
|
|
|
END_C_DECLS
|
|
|
|
#endif
|