1
1
openmpi/opal/mca/pmix/cray/pmix_cray.c

845 строки
27 KiB
C
Исходник Обычный вид История

Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2007 The Trustees of Indiana University.
* All rights reserved.
* Copyright (c) 2011 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2011-2015 Los Alamos National Security, LLC. All
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
* rights reserved.
* Copyright (c) 2013-2015 Intel, Inc. All rights reserved.
* Copyright (c) 2014 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "opal_config.h"
#include "opal/constants.h"
#include "opal/types.h"
#include "opal_stdint.h"
#include "opal/mca/hwloc/base/base.h"
#include "opal/util/opal_environ.h"
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
#include "opal/util/output.h"
#include "opal/util/proc.h"
#include "opal/util/output.h"
#include "opal/util/show_help.h"
#include <string.h>
#include <pmi.h>
#include <pmi2.h>
#include "opal/mca/pmix/base/base.h"
#include "opal/mca/pmix/base/pmix_base_hash.h"
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
#include "pmix_cray.h"
static char cray_pmi_version[128];
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
static int cray_init(void);
static int cray_fini(void);
static int cray_initialized(void);
static int cray_abort(int flat, const char *msg,
opal_list_t *procs);
2015-08-29 17:26:32 -07:00
static int cray_spawn(opal_list_t *jobinfo, opal_list_t *apps, opal_jobid_t *jobid);
static int cray_spawn_nb(opal_list_t *jobinfo, opal_list_t *apps,
opal_pmix_spawn_cbfunc_t cbfunc,
void *cbdata);
static int cray_job_connect(opal_list_t *procs);
static int cray_job_disconnect(opal_list_t *procs);
static int cray_job_disconnect_nb(opal_list_t *procs,
opal_pmix_op_cbfunc_t cbfunc,
void *cbdata);
static int cray_resolve_peers(const char *nodename,
opal_jobid_t jobid,
opal_list_t *procs);
static int cray_resolve_nodes(opal_jobid_t jobid, char **nodelist);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
static int cray_put(opal_pmix_scope_t scope, opal_value_t *kv);
static int cray_fence(opal_list_t *procs, int collect_data);
static int cray_fence_nb(opal_list_t *procs, int collect_data,
opal_pmix_op_cbfunc_t cbfunc, void *cbdata);
static int cray_commit(void);
static int cray_get(const opal_process_name_t *id,
const char *key, opal_list_t *info,
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
opal_value_t **kv);
static int cray_get_nb(const opal_process_name_t *id, const char *key,
opal_list_t *info,
opal_pmix_value_cbfunc_t cbfunc, void *cbdata);
static int cray_publish(opal_list_t *info);
static int cray_publish_nb(opal_list_t *info,
opal_pmix_op_cbfunc_t cbfunc, void *cbdata);
static int cray_lookup(opal_list_t *data, opal_list_t *info);
static int cray_lookup_nb(char **keys, opal_list_t *info,
opal_pmix_lookup_cbfunc_t cbfunc, void *cbdata);
static int cray_unpublish(char **keys, opal_list_t *info);
static int cray_unpublish_nb(char **keys, opal_list_t *info,
opal_pmix_op_cbfunc_t cbfunc, void *cbdata);
static const char *cray_get_version(void);
static int cray_store_local(const opal_process_name_t *proc,
opal_value_t *val);
#if 0
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
static bool cray_get_attr(const char *attr, opal_value_t **kv);
#endif
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
const opal_pmix_base_module_t opal_pmix_cray_module = {
.init = cray_init,
.finalize = cray_fini,
.initialized = cray_initialized,
.abort = cray_abort,
.commit = cray_commit,
.fence = cray_fence,
.fence_nb = cray_fence_nb,
.put = cray_put,
.get = cray_get,
.get_nb = cray_get_nb,
.publish = cray_publish,
.publish_nb = cray_publish_nb,
.lookup = cray_lookup,
.lookup_nb = cray_lookup_nb,
.unpublish = cray_unpublish,
.unpublish_nb = cray_unpublish_nb,
.spawn = cray_spawn,
.spawn_nb = cray_spawn_nb,
.connect = cray_job_connect,
.disconnect = cray_job_disconnect,
.disconnect_nb = cray_job_disconnect_nb,
.resolve_peers = cray_resolve_peers,
.resolve_nodes = cray_resolve_nodes,
.get_version = cray_get_version,
.register_errhandler = opal_pmix_base_register_handler,
.deregister_errhandler = opal_pmix_base_deregister_handler,
.store_local = cray_store_local
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
};
// usage accounting
static int pmix_init_count = 0;
// PMI constant values:
static int pmix_kvslen_max = 0;
static int pmix_keylen_max = 0;
static int pmix_vallen_max = 0;
static int pmix_vallen_threshold = INT_MAX;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
// Job environment description
static int pmix_size = 0;
static int pmix_rank = 0;
static int pmix_lrank = 0;
static int pmix_nrank = 0;
static int pmix_nlranks = 0;
static int pmix_appnum = 0;
static int pmix_usize = 0;
static char *pmix_kvs_name = NULL;
static int *pmix_lranks = NULL;
static opal_process_name_t pmix_pname;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
static uint32_t pmix_jobid = -1;
static char* pmix_error(int pmix_err);
#define OPAL_PMI_ERROR(pmi_err, pmi_func) \
do { \
opal_output(0, "%s [%s:%d:%s]: %s\n", \
pmi_func, __FILE__, __LINE__, __func__, \
pmix_error(pmi_err)); \
} while(0);
static int cray_init(void)
{
int i, spawned, size, rank, appnum, my_node;
int rc, ret = OPAL_ERROR;
char *pmapping = NULL;
char buf[PMI2_MAX_ATTRVALUE];
int found;
int major, minor, revision;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
uint32_t jobfam;
opal_value_t kv;
opal_process_name_t ldr;
char nmtmp[64];
char *str, **localranks = NULL;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
++pmix_init_count;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
/* if we can't startup PMI, we can't be used */
if ( PMI2_Initialized () ) {
opal_output_verbose(10, opal_pmix_base_framework.framework_output,
"%s pmix:cray: pmi already initialized",
OPAL_NAME_PRINT(pmix_pname));
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
return OPAL_SUCCESS;
}
size = -1;
rank = -1;
appnum = -1;
if (PMI_SUCCESS != (rc = PMI2_Init(&spawned, &size, &rank, &appnum))) {
opal_show_help("help-pmix-base.txt", "pmix2-init-failed", true, rc);
return OPAL_ERROR;
}
if( size < 0 || rank < 0 ){
opal_show_help("help-pmix-base.txt", "pmix2-init-returned-bad-values", true);
goto err_exit;
}
pmix_size = size;
pmix_rank = rank;
pmix_appnum = appnum;
pmix_vallen_max = PMI2_MAX_VALLEN;
pmix_kvslen_max = PMI2_MAX_VALLEN; // FIX ME: What to put here for versatility?
pmix_keylen_max = PMI2_MAX_KEYLEN;
pmix_vallen_threshold = PMI2_MAX_VALLEN * 3;
pmix_vallen_threshold >>= 2;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
/*
* get the version info
*/
if (PMI_SUCCESS != PMI_Get_version_info(&major,&minor,&revision)) {
return OPAL_ERROR;
}
snprintf(cray_pmi_version, sizeof(cray_pmi_version),
"%d.%d.%d", major, minor, revision);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
pmix_kvs_name = (char*)malloc(pmix_kvslen_max);
if( pmix_kvs_name == NULL ){
PMI2_Finalize();
ret = OPAL_ERR_OUT_OF_RESOURCE;
goto err_exit;
}
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
rc = PMI2_Job_GetId(pmix_kvs_name, pmix_kvslen_max);
if( PMI_SUCCESS != rc ) {
OPAL_PMI_ERROR(rc, "PMI2_Job_GetId");
goto err_exit;
}
rc = sscanf(pmix_kvs_name,"kvs_%u",&jobfam);
if (rc != 1) {
opal_output_verbose(10, opal_pmix_base_framework.framework_output,
"%s pmix:cray: pmix_kvs_name %s",
OPAL_NAME_PRINT(pmix_pname), pmix_kvs_name);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
rc = OPAL_ERROR;
goto err_exit;
}
pmix_jobid = jobfam << 16;
/* store our name in the opal_proc_t so that
* debug messages will make sense - an upper
* layer will eventually overwrite it, but that
* won't do any harm */
pmix_pname.jobid = pmix_jobid;
pmix_pname.vpid = pmix_rank;
opal_proc_set_name(&pmix_pname);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
opal_output_verbose(10, opal_pmix_base_framework.framework_output,
"%s pmix:cray: assigned tmp name %d %d pmix_kvs_name %s",
OPAL_NAME_PRINT(pmix_pname),pmix_pname.jobid,pmix_pname.vpid,pmix_kvs_name);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
pmapping = (char*)malloc(PMI2_MAX_VALLEN);
if( pmapping == NULL ){
rc = OPAL_ERR_OUT_OF_RESOURCE;
OPAL_ERROR_LOG(rc);
return rc;
}
rc = PMI2_Info_GetJobAttr("PMI_process_mapping", pmapping, PMI2_MAX_VALLEN, &found);
if( !found || PMI_SUCCESS != rc ) {
OPAL_PMI_ERROR(rc,"PMI2_Info_GetJobAttr");
return OPAL_ERROR;
}
pmix_lranks = pmix_cray_parse_pmap(pmapping, pmix_rank, &my_node, &pmix_nlranks);
if (NULL == pmix_lranks) {
rc = OPAL_ERR_OUT_OF_RESOURCE;
OPAL_ERROR_LOG(rc);
return rc;
}
free(pmapping);
// setup hash table
opal_pmix_base_hash_init();
/* save the job size */
OBJ_CONSTRUCT(&kv, opal_value_t);
kv.key = strdup(OPAL_PMIX_JOB_SIZE);
kv.type = OPAL_UINT32;
kv.data.uint32 = pmix_size;
if (OPAL_SUCCESS != (rc = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) {
OPAL_ERROR_LOG(rc);
OBJ_DESTRUCT(&kv);
goto err_exit;
}
OBJ_DESTRUCT(&kv);
/* save the appnum */
OBJ_CONSTRUCT(&kv, opal_value_t);
kv.key = strdup(OPAL_PMIX_APPNUM);
kv.type = OPAL_UINT32;
kv.data.uint32 = pmix_appnum;
if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) {
OPAL_ERROR_LOG(ret);
OBJ_DESTRUCT(&kv);
goto err_exit;
}
OBJ_DESTRUCT(&kv);
rc = PMI2_Info_GetJobAttr("universeSize", buf, 16, &found);
if( PMI_SUCCESS != rc ) {
OPAL_PMI_ERROR(rc, "PMI_Get_universe_size");
goto err_exit;
}
pmix_usize = atoi(buf);
OBJ_CONSTRUCT(&kv, opal_value_t);
kv.key = strdup(OPAL_PMIX_UNIV_SIZE);
kv.type = OPAL_UINT32;
kv.data.uint32 = pmix_usize;
if (OPAL_SUCCESS != (rc = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) {
OPAL_ERROR_LOG(rc);
OBJ_DESTRUCT(&kv);
goto err_exit;
}
OBJ_DESTRUCT(&kv);
OBJ_CONSTRUCT(&kv, opal_value_t);
kv.key = strdup(OPAL_PMIX_JOBID);
kv.type = OPAL_UINT32;
kv.data.uint32 = pmix_jobid;
if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) {
OPAL_ERROR_LOG(ret);
OBJ_DESTRUCT(&kv);
goto err_exit;
}
OBJ_DESTRUCT(&kv);
/* save the local size */
OBJ_CONSTRUCT(&kv, opal_value_t);
kv.key = strdup(OPAL_PMIX_LOCAL_SIZE);
kv.type = OPAL_UINT32;
kv.data.uint32 = pmix_nlranks;
if (OPAL_SUCCESS != (rc = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) {
OPAL_ERROR_LOG(rc);
OBJ_DESTRUCT(&kv);
goto err_exit;
}
OBJ_DESTRUCT(&kv);
ldr.vpid = pmix_lranks[0];
ldr.jobid = pmix_pname.jobid;
/* find ourselves and build up a string for local peer info */
memset(nmtmp, 0, 64);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
for (i=0; i < pmix_nlranks; i++) {
ret = snprintf(nmtmp, 64, "%d", pmix_lranks[i]);
opal_argv_append_nosize(&localranks, nmtmp);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
if (pmix_rank == pmix_lranks[i]) {
pmix_lrank = i;
pmix_nrank = i;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
}
str = opal_argv_join(localranks, ',');
opal_argv_free(localranks);
OBJ_CONSTRUCT(&kv, opal_value_t);
kv.key = strdup(OPAL_PMIX_LOCAL_PEERS);
kv.type = OPAL_STRING;
kv.data.string = str;
if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) {
OPAL_ERROR_LOG(ret);
OBJ_DESTRUCT(&kv);
goto err_exit;
}
OBJ_DESTRUCT(&kv);
/* save the local leader */
OBJ_CONSTRUCT(&kv, opal_value_t);
kv.key = strdup(OPAL_PMIX_LOCALLDR);
kv.type = OPAL_UINT64;
kv.data.uint64 = *(uint64_t*)&ldr;
if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) {
OPAL_ERROR_LOG(ret);
OBJ_DESTRUCT(&kv);
goto err_exit;
}
/* save our local rank */
OBJ_CONSTRUCT(&kv, opal_value_t);
kv.key = strdup(OPAL_PMIX_LOCAL_RANK);
kv.type = OPAL_UINT16;
kv.data.uint16 = pmix_lrank;
if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) {
OPAL_ERROR_LOG(ret);
OBJ_DESTRUCT(&kv);
goto err_exit;
}
/* and our node rank */
OBJ_CONSTRUCT(&kv, opal_value_t);
kv.key = strdup(OPAL_PMIX_NODE_RANK);
kv.type = OPAL_UINT16;
kv.data.uint16 = pmix_nrank;
if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) {
OPAL_ERROR_LOG(ret);
OBJ_DESTRUCT(&kv);
goto err_exit;
}
OBJ_DESTRUCT(&kv);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
return OPAL_SUCCESS;
err_exit:
PMI2_Finalize();
return ret;
}
static int cray_fini(void) {
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
if (0 == pmix_init_count) {
return OPAL_SUCCESS;
}
if (0 == --pmix_init_count) {
PMI2_Finalize();
}
if (NULL != pmix_kvs_name) {
free(pmix_kvs_name);
pmix_kvs_name = NULL;
}
if (NULL != pmix_lranks) {
free(pmix_lranks);
pmix_lranks = NULL;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
return OPAL_SUCCESS;
}
static int cray_initialized(void)
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
{
if (0 < pmix_init_count) {
return 1;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
return 0;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
static int cray_abort(int flag, const char *msg,
opal_list_t *procs)
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
{
PMI2_Abort(flag, msg);
return OPAL_SUCCESS;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
2015-08-29 17:26:32 -07:00
static int cray_spawn(opal_list_t *jobinfo, opal_list_t *apps, opal_jobid_t *jobid)
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
{
return OPAL_ERR_NOT_SUPPORTED;
}
static int cray_spawn_nb(opal_list_t *jobinfo, opal_list_t *apps,
opal_pmix_spawn_cbfunc_t cbfunc,
void *cbdata)
{
return OPAL_ERR_NOT_SUPPORTED;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
static int cray_job_connect(opal_list_t *procs)
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
{
return OPAL_ERR_NOT_SUPPORTED;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
static int cray_job_disconnect(opal_list_t *procs)
{
return OPAL_ERR_NOT_SUPPORTED;
}
static int cray_job_disconnect_nb(opal_list_t *procs,
opal_pmix_op_cbfunc_t cbfunc,
void *cbdata)
{
return OPAL_ERR_NOT_SUPPORTED;
}
static int cray_resolve_peers(const char *nodename,
opal_jobid_t jobid,
opal_list_t *procs)
{
return OPAL_ERR_NOT_IMPLEMENTED;
}
static int cray_resolve_nodes(opal_jobid_t jobid, char **nodelist)
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
{
return OPAL_ERR_NOT_IMPLEMENTED;
}
static int cray_put(opal_pmix_scope_t scope,
opal_value_t *kv)
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
{
int rc;
opal_output_verbose(10, opal_pmix_base_framework.framework_output,
"%s pmix:cray cray_put key %s scope %d\n",
OPAL_NAME_PRINT(OPAL_PROC_MY_NAME), kv->key, scope);
if (!pmix_init_count) {
return OPAL_ERROR;
}
/*
* for now just always just global cache
*/
if (NULL == mca_pmix_cray_component.cache_global) {
mca_pmix_cray_component.cache_global = OBJ_NEW(opal_buffer_t);
}
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
opal_output_verbose(20, opal_pmix_base_framework.framework_output,
"%s pmix:cray put global data for key %s type %d",
OPAL_NAME_PRINT(OPAL_PROC_MY_NAME), kv->key, kv->type);
if (OPAL_SUCCESS != (rc = opal_dss.pack(mca_pmix_cray_component.cache_global, &kv, 1, OPAL_VALUE))) {
OPAL_PMI_ERROR(rc,"pmix:cray opal_dss.pack returned error");
OPAL_ERROR_LOG(rc);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
return rc;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
static int cray_commit(void)
{
return OPAL_SUCCESS;
}
static int cray_fence(opal_list_t *procs, int collect_data)
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
{
int rc, cnt;
int32_t i;
int *all_lens = NULL;
opal_value_t *kp, kvn;
opal_buffer_t *send_buffer = NULL;
opal_buffer_t *buf = NULL;
void *sbuf_ptr;
char *cptr, *rcv_buff = NULL;
opal_process_name_t id;
typedef struct {
uint32_t pmix_rank;
opal_process_name_t name;
int32_t nbytes;
} bytes_and_rank_t;
int32_t rcv_nbytes_tot;
bytes_and_rank_t s_bytes_and_rank;
bytes_and_rank_t *r_bytes_and_ranks = NULL;
opal_hwloc_locality_t locality;
opal_list_t vals;
char *cpuset = NULL;
opal_process_name_t pname;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
opal_output_verbose(2, opal_pmix_base_framework.framework_output,
"%s pmix:cray executing fence cache_global %p cache_local %p",
OPAL_NAME_PRINT(OPAL_PROC_MY_NAME),
(void *)mca_pmix_cray_component.cache_global,
(void *)mca_pmix_cray_component.cache_local);
/* get the modex data from each local process and set the
* localities to avoid having the MPI layer fetch data
* for every process in the job */
pname.jobid = OPAL_PROC_MY_NAME.jobid;
/*
* "unload" the cache_local/cache_global buffers, first copy
* it so we can continue to use the local buffers if further
* calls to put can be made
*/
send_buffer = OBJ_NEW(opal_buffer_t);
if (NULL == send_buffer) {
return OPAL_ERR_OUT_OF_RESOURCE;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
opal_dss.copy_payload(send_buffer, mca_pmix_cray_component.cache_global);
opal_dss.unload(send_buffer, &sbuf_ptr, &s_bytes_and_rank.nbytes);
s_bytes_and_rank.pmix_rank = pmix_rank;
s_bytes_and_rank.name = OPAL_PROC_MY_NAME;
r_bytes_and_ranks = (bytes_and_rank_t *)malloc(pmix_size * sizeof(bytes_and_rank_t));
if (NULL == r_bytes_and_ranks) {
rc = OPAL_ERR_OUT_OF_RESOURCE;
goto fn_exit;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
/*
* gather up all the buffer sizes and rank order.
* doing this step below since the cray pmi PMI_Allgather doesn't deliver
* the gathered data necessarily in PMI rank order, although the order stays
* the same for the duration of a job - assuming no node failures.
*/
if (PMI_SUCCESS != (rc = PMI_Allgather(&s_bytes_and_rank,r_bytes_and_ranks,sizeof(bytes_and_rank_t)))) {
OPAL_PMI_ERROR(rc,"PMI_Allgather");
rc = OPAL_ERR_COMM_FAILURE;
goto fn_exit;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
for (rcv_nbytes_tot=0,i=0; i < pmix_size; i++) {
rcv_nbytes_tot += r_bytes_and_ranks[i].nbytes;
}
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
opal_output_verbose(20, opal_pmix_base_framework.framework_output,
"%s pmix:cray total number of bytes to receive %d",
OPAL_NAME_PRINT(OPAL_PROC_MY_NAME), rcv_nbytes_tot);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
rcv_buff = (char *) malloc(rcv_nbytes_tot * sizeof(char));
if (NULL == rcv_buff) {
rc = OPAL_ERR_OUT_OF_RESOURCE;
goto fn_exit;
}
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
all_lens = (int *)malloc(sizeof(int) * pmix_size);
if (NULL == all_lens) {
rc = OPAL_ERR_OUT_OF_RESOURCE;
goto fn_exit;
}
for (i=0; i< pmix_size; i++) {
all_lens[r_bytes_and_ranks[i].pmix_rank] = r_bytes_and_ranks[i].nbytes;
}
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
if (PMI_SUCCESS != (rc = PMI_Allgatherv(sbuf_ptr,s_bytes_and_rank.nbytes,rcv_buff,all_lens))) {
OPAL_PMI_ERROR(rc,"PMI_Allgatherv");
rc = OPAL_ERR_COMM_FAILURE;
goto fn_exit;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
OBJ_RELEASE(send_buffer);
send_buffer = NULL;
buf = OBJ_NEW(opal_buffer_t);
if (buf == NULL) {
rc = OPAL_ERR_OUT_OF_RESOURCE;
goto fn_exit;
}
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
for (cptr = rcv_buff, i=0; i < pmix_size; i++) {
id = r_bytes_and_ranks[i].name;
buf->base_ptr = NULL; /* TODO: ugh */
if (OPAL_SUCCESS != (rc = opal_dss.load(buf, (void *)cptr, r_bytes_and_ranks[i].nbytes))) {
OPAL_PMI_ERROR(rc,"pmix:cray opal_dss.load failed");
goto fn_exit;
}
/* unpack and stuff in to the dstore */
cnt = 1;
while (OPAL_SUCCESS == (rc = opal_dss.unpack(buf, &kp, &cnt, OPAL_VALUE))) {
opal_output_verbose(20, opal_pmix_base_framework.framework_output,
2015-06-23 20:59:57 -07:00
"%s pmix:cray unpacked kp with key %s type(%d) for id %s",
OPAL_NAME_PRINT(OPAL_PROC_MY_NAME), kp->key, kp->type, OPAL_NAME_PRINT(id));
if (OPAL_SUCCESS != (rc = opal_pmix_base_store(&id, kp))) {
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
OPAL_ERROR_LOG(rc);
goto fn_exit;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
OBJ_RELEASE(kp);
cnt = 1;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
cptr += r_bytes_and_ranks[i].nbytes;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
buf->base_ptr = NULL; /* TODO: ugh */
OBJ_RELEASE(buf);
opal_output_verbose(2, opal_pmix_base_framework.framework_output,
"%s pmix:cray kvs_fence complete",
OPAL_NAME_PRINT(OPAL_PROC_MY_NAME));
/* fetch my cpuset */
OBJ_CONSTRUCT(&vals, opal_list_t);
if (OPAL_SUCCESS == (rc = opal_pmix_base_fetch(&pmix_pname,
OPAL_PMIX_CPUSET, &vals))) {
kp = (opal_value_t*)opal_list_get_first(&vals);
cpuset = strdup(kp->data.string);
} else {
cpuset = NULL;
}
OPAL_LIST_DESTRUCT(&vals);
/* we only need to set locality for each local rank as "not found"
* equates to "non-local" */
for (i=0; i < pmix_nlranks; i++) {
id.vpid = pmix_lranks[i];
id.jobid = pmix_jobid;
opal_output_verbose(2, opal_pmix_base_framework.framework_output,
"%s checking out if %s is local to me",
OPAL_NAME_PRINT(OPAL_PROC_MY_NAME),
OPAL_NAME_PRINT(id));
/* fetch cpuset for this vpid */
OBJ_CONSTRUCT(&vals, opal_list_t);
if (OPAL_SUCCESS != (rc = opal_pmix_base_fetch(&id,
OPAL_PMIX_CPUSET, &vals))) {
opal_output_verbose(2, opal_pmix_base_framework.framework_output,
"%s cpuset for local proc %s not found",
OPAL_NAME_PRINT(OPAL_PROC_MY_NAME),
OPAL_NAME_PRINT(id));
OPAL_LIST_DESTRUCT(&vals);
/* even though the cpuset wasn't found, we at least know it is
* on the same node with us */
locality = OPAL_PROC_ON_CLUSTER | OPAL_PROC_ON_CU | OPAL_PROC_ON_NODE;
} else {
kp = (opal_value_t*)opal_list_get_first(&vals);
if (NULL == kp->data.string) {
/* if we share a node, but we don't know anything more, then
* mark us as on the node as this is all we know
*/
locality = OPAL_PROC_ON_CLUSTER | OPAL_PROC_ON_CU | OPAL_PROC_ON_NODE;
} else {
/* determine relative location on our node */
locality = opal_hwloc_base_get_relative_locality(opal_hwloc_topology,
cpuset,
kp->data.string);
}
OPAL_LIST_DESTRUCT(&vals);
}
OPAL_OUTPUT_VERBOSE((1, opal_pmix_base_framework.framework_output,
"%s pmix:cray proc %s locality %s",
OPAL_NAME_PRINT(OPAL_PROC_MY_NAME),
OPAL_NAME_PRINT(id),
opal_hwloc_base_print_locality(locality)));
OBJ_CONSTRUCT(&kvn, opal_value_t);
kvn.key = strdup(OPAL_PMIX_LOCALITY);
kvn.type = OPAL_UINT16;
kvn.data.uint16 = locality;
opal_pmix_base_store(&pname, &kvn);
OBJ_DESTRUCT(&kvn);
}
fn_exit:
if (NULL != cpuset) {
free(cpuset);
}
if (all_lens != NULL) {
free(all_lens);
}
if (rcv_buff != NULL) {
free(rcv_buff);
}
if (r_bytes_and_ranks != NULL) {
free(r_bytes_and_ranks);
}
return rc;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
static int cray_fence_nb(opal_list_t *procs, int collect_data,
opal_pmix_op_cbfunc_t cbfunc, void *cbdata)
{
return OPAL_ERR_NOT_IMPLEMENTED;
}
static int cray_get(const opal_process_name_t *id, const char *key, opal_list_t *info, opal_value_t **kv)
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
{
int rc;
opal_list_t vals;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
opal_output_verbose(2, opal_pmix_base_framework.framework_output,
"%s pmix:cray getting value for proc %s key %s",
OPAL_NAME_PRINT(OPAL_PROC_MY_NAME),
OPAL_NAME_PRINT(*id), key);
OBJ_CONSTRUCT(&vals, opal_list_t);
rc = opal_pmix_base_fetch(id, key, &vals);
if (OPAL_SUCCESS == rc) {
*kv = (opal_value_t*)opal_list_remove_first(&vals);
return OPAL_SUCCESS;
} else {
opal_output_verbose(2, opal_pmix_base_framework.framework_output,
"%s pmix:cray fetch from dstore failed: %d",
OPAL_NAME_PRINT(OPAL_PROC_MY_NAME), rc);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
OPAL_LIST_DESTRUCT(&vals);
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
return rc;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
static int cray_get_nb(const opal_process_name_t *id, const char *key,
opal_list_t *info, opal_pmix_value_cbfunc_t cbfunc, void *cbdata)
{
return OPAL_ERR_NOT_IMPLEMENTED;
}
static int cray_publish(opal_list_t *info)
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
{
return OPAL_ERR_NOT_SUPPORTED;
}
static int cray_publish_nb(opal_list_t *info,
opal_pmix_op_cbfunc_t cbfunc, void *cbdata)
{
return OPAL_ERR_NOT_SUPPORTED;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
static int cray_lookup(opal_list_t *data, opal_list_t *info)
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
{
return OPAL_ERR_NOT_SUPPORTED;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
static int cray_lookup_nb(char **keys, opal_list_t *info,
opal_pmix_lookup_cbfunc_t cbfunc, void *cbdata)
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
{
return OPAL_ERR_NOT_SUPPORTED;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
static int cray_unpublish(char **keys, opal_list_t *info)
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
{
return OPAL_ERR_NOT_SUPPORTED;
}
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
static int cray_unpublish_nb(char **keys, opal_list_t *info,
opal_pmix_op_cbfunc_t cbfunc, void *cbdata)
{
return OPAL_ERR_NOT_SUPPORTED;
}
static const char *cray_get_version(void)
{
return cray_pmi_version;
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
}
static int cray_store_local(const opal_process_name_t *proc,
opal_value_t *val)
{
opal_pmix_base_store(proc, val);
return OPAL_SUCCESS;
}
Per the PMIx RFC: WHAT: Merge the PMIx branch into the devel repo, creating a new OPAL “lmix” framework to abstract PMI support for all RTEs. Replace the ORTE daemon-level collectives with a new PMIx server and update the ORTE grpcomm framework to support server-to-server collectives WHY: We’ve had problems dealing with variations in PMI implementations, and need to extend the existing PMI definitions to meet exascale requirements. WHEN: Mon, Aug 25 WHERE: https://github.com/rhc54/ompi-svn-mirror.git Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding. All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level. Accordingly, we have: * created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations. * Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported. * Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint * removed the prior OMPI/OPAL modex code * added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform. * retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
static char* pmix_error(int pmix_err)
{
char * err_msg;
switch(pmix_err) {
case PMI_FAIL: err_msg = "Operation failed"; break;
case PMI_ERR_INIT: err_msg = "PMI is not initialized"; break;
case PMI_ERR_NOMEM: err_msg = "Input buffer not large enough"; break;
case PMI_ERR_INVALID_ARG: err_msg = "Invalid argument"; break;
case PMI_ERR_INVALID_KEY: err_msg = "Invalid key argument"; break;
case PMI_ERR_INVALID_KEY_LENGTH: err_msg = "Invalid key length argument"; break;
case PMI_ERR_INVALID_VAL: err_msg = "Invalid value argument"; break;
case PMI_ERR_INVALID_VAL_LENGTH: err_msg = "Invalid value length argument"; break;
case PMI_ERR_INVALID_LENGTH: err_msg = "Invalid length argument"; break;
case PMI_ERR_INVALID_NUM_ARGS: err_msg = "Invalid number of arguments"; break;
case PMI_ERR_INVALID_ARGS: err_msg = "Invalid args argument"; break;
case PMI_ERR_INVALID_NUM_PARSED: err_msg = "Invalid num_parsed length argument"; break;
case PMI_ERR_INVALID_KEYVALP: err_msg = "Invalid keyvalp argument"; break;
case PMI_ERR_INVALID_SIZE: err_msg = "Invalid size argument"; break;
#if defined(PMI_ERR_INVALID_KVS)
/* pmi.h calls this a valid return code but mpich doesn't define it (slurm does). */
case PMI_ERR_INVALID_KVS: err_msg = "Invalid kvs argument"; break;
#endif
case PMI_SUCCESS: err_msg = "Success"; break;
default: err_msg = "Unkown error";
}
return err_msg;
}