ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность
cd72024a3c
openmpi/config/opal_check_pmi.m4

147 строки
6.9 KiB
Plaintext
Исходник Обычный вид История

first cut at general pmi check This commit was SVN r25234.
2011-10-05 21:14:24 +04:00
# -*- shell-script -*-
#
# Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
# University Research and Technology
# Corporation. All rights reserved.
# Copyright (c) 2004-2005 The University of Tennessee and The University
# of Tennessee Research Foundation. All rights
# reserved.
update ess alps configuration file to include CNOS and PMI checks. some of the features committed here aren't being used, but they will be. also update orte_check_pmi.m4 to include missing call to action-if-not-found if --with-pmi is not specified or is disabled. This commit was SVN r25422.
2011-11-03 06:14:47 +04:00
# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
first cut at general pmi check This commit was SVN r25234.
2011-10-05 21:14:24 +04:00
# University of Stuttgart. All rights reserved.
# Copyright (c) 2004-2005 The Regents of the University of California.
# All rights reserved.
Complete implementation of pmi support. Ensure we support both mpirun and direct launch within same configuration to avoid requiring separate builds. Add support for generic pmi, not just under slurm. Add publish/subscribe support, although slurm's pmi implementation will just return an error as it hasn't been done yet. This commit was SVN r25303.
2011-10-18 00:51:22 +04:00
# Copyright (c) 2009-2011 Cisco Systems, Inc. All rights reserved.
Per the meeting on moving the BTLs to OPAL, move the ORTE database "db" framework to OPAL so the relocated BTLs can access it. Because the data is indexed by process, this requires that we define a new "opal_identifier_t" that corresponds to the orte_process_name_t struct. In order to support multiple run-times, this is defined in opal/mca/db/db_types.h as a uint64_t without identifying the meaning of any part of that data. A few changes were required to support this move: 1. the PMI component used to identify rte-related data (e.g., host name, bind level) and package them as a unit to reduce the number of PMI keys. This code was moved up to the ORTE layer as the OPAL layer has no understanding of these concepts. In addition, the component locally stored data based on process jobid/vpid - this could no longer be supported (see below for the solution). 2. the hash component was updated to use the new opal_identifier_t instead of orte_process_name_t as its index for storing data in the hash tables. Previously, we did a hash on the vpid and stored the data in a 32-bit hash table. In the revised system, we don't see a separate "vpid" field - we only have a 64-bit opaque value. The orte_process_name_t hash turned out to do nothing useful, so we now store the data in a 64-bit hash table. Preliminary tests didn't show any identifiable change in behavior or performance, but we'll have to see if a move back to the 32-bit table is required at some later time. 3. the db framework was a "select one" system. However, since the PMI component could no longer use its internal storage system, the framework has now been changed to a "select many" mode of operation. This allows the hash component to handle all internal storage, while the PMI component only handles pushing/pulling things from the PMI system. This was something we had planned for some time - when fetching data, we first check internal storage to see if we already have it, and then automatically go to the global system to look for it if we don't. Accordingly, the framework was provided with a custom query function used during "select" that lets you seperately specify the "store" and "fetch" ordering. 4. the ORTE grpcomm and ess/pmi components, and the nidmap code, were updated to work with the new db framework and to specify internal/global storage options. No changes were made to the MPI layer, except for modifying the ORTE component of the OMPI/rte framework to support the new db framework. This commit was SVN r28112.
2013-02-26 21:50:04 +04:00
# Copyright (c) 2011-2013 Los Alamos National Security, LLC. All rights
first cut at general pmi check This commit was SVN r25234.
2011-10-05 21:14:24 +04:00
# reserved.
# $COPYRIGHT$
update ess alps configuration file to include CNOS and PMI checks. some of the features committed here aren't being used, but they will be. also update orte_check_pmi.m4 to include missing call to action-if-not-found if --with-pmi is not specified or is disabled. This commit was SVN r25422.
2011-11-03 06:14:47 +04:00
#
first cut at general pmi check This commit was SVN r25234.
2011-10-05 21:14:24 +04:00
# Additional copyrights may follow
update ess alps configuration file to include CNOS and PMI checks. some of the features committed here aren't being used, but they will be. also update orte_check_pmi.m4 to include missing call to action-if-not-found if --with-pmi is not specified or is disabled. This commit was SVN r25422.
2011-11-03 06:14:47 +04:00
#
first cut at general pmi check This commit was SVN r25234.
2011-10-05 21:14:24 +04:00
# $HEADER$
#
Per the meeting on moving the BTLs to OPAL, move the ORTE database "db" framework to OPAL so the relocated BTLs can access it. Because the data is indexed by process, this requires that we define a new "opal_identifier_t" that corresponds to the orte_process_name_t struct. In order to support multiple run-times, this is defined in opal/mca/db/db_types.h as a uint64_t without identifying the meaning of any part of that data. A few changes were required to support this move: 1. the PMI component used to identify rte-related data (e.g., host name, bind level) and package them as a unit to reduce the number of PMI keys. This code was moved up to the ORTE layer as the OPAL layer has no understanding of these concepts. In addition, the component locally stored data based on process jobid/vpid - this could no longer be supported (see below for the solution). 2. the hash component was updated to use the new opal_identifier_t instead of orte_process_name_t as its index for storing data in the hash tables. Previously, we did a hash on the vpid and stored the data in a 32-bit hash table. In the revised system, we don't see a separate "vpid" field - we only have a 64-bit opaque value. The orte_process_name_t hash turned out to do nothing useful, so we now store the data in a 64-bit hash table. Preliminary tests didn't show any identifiable change in behavior or performance, but we'll have to see if a move back to the 32-bit table is required at some later time. 3. the db framework was a "select one" system. However, since the PMI component could no longer use its internal storage system, the framework has now been changed to a "select many" mode of operation. This allows the hash component to handle all internal storage, while the PMI component only handles pushing/pulling things from the PMI system. This was something we had planned for some time - when fetching data, we first check internal storage to see if we already have it, and then automatically go to the global system to look for it if we don't. Accordingly, the framework was provided with a custom query function used during "select" that lets you seperately specify the "store" and "fetch" ordering. 4. the ORTE grpcomm and ess/pmi components, and the nidmap code, were updated to work with the new db framework and to specify internal/global storage options. No changes were made to the MPI layer, except for modifying the ORTE component of the OMPI/rte framework to support the new db framework. This commit was SVN r28112.
2013-02-26 21:50:04 +04:00
# OPAL_CHECK_PMI(prefix, [action-if-found], [action-if-not-found])
first cut at general pmi check This commit was SVN r25234.
2011-10-05 21:14:24 +04:00
# --------------------------------------------------------
Per the meeting on moving the BTLs to OPAL, move the ORTE database "db" framework to OPAL so the relocated BTLs can access it. Because the data is indexed by process, this requires that we define a new "opal_identifier_t" that corresponds to the orte_process_name_t struct. In order to support multiple run-times, this is defined in opal/mca/db/db_types.h as a uint64_t without identifying the meaning of any part of that data. A few changes were required to support this move: 1. the PMI component used to identify rte-related data (e.g., host name, bind level) and package them as a unit to reduce the number of PMI keys. This code was moved up to the ORTE layer as the OPAL layer has no understanding of these concepts. In addition, the component locally stored data based on process jobid/vpid - this could no longer be supported (see below for the solution). 2. the hash component was updated to use the new opal_identifier_t instead of orte_process_name_t as its index for storing data in the hash tables. Previously, we did a hash on the vpid and stored the data in a 32-bit hash table. In the revised system, we don't see a separate "vpid" field - we only have a 64-bit opaque value. The orte_process_name_t hash turned out to do nothing useful, so we now store the data in a 64-bit hash table. Preliminary tests didn't show any identifiable change in behavior or performance, but we'll have to see if a move back to the 32-bit table is required at some later time. 3. the db framework was a "select one" system. However, since the PMI component could no longer use its internal storage system, the framework has now been changed to a "select many" mode of operation. This allows the hash component to handle all internal storage, while the PMI component only handles pushing/pulling things from the PMI system. This was something we had planned for some time - when fetching data, we first check internal storage to see if we already have it, and then automatically go to the global system to look for it if we don't. Accordingly, the framework was provided with a custom query function used during "select" that lets you seperately specify the "store" and "fetch" ordering. 4. the ORTE grpcomm and ess/pmi components, and the nidmap code, were updated to work with the new db framework and to specify internal/global storage options. No changes were made to the MPI layer, except for modifying the ORTE component of the OMPI/rte framework to support the new db framework. This commit was SVN r28112.
2013-02-26 21:50:04 +04:00
AC_DEFUN([OPAL_CHECK_PMI],[
Complete the PMI support. Generalize PMI operations to support both slurm and non-slurm environments. Correct some configuration issues - we really only want the PMI integration at the individual component level. Ensure that the pmi grpcomm component doesn't get selected when launching via mpirun by setting its priority below the bad component. Only verified in a slurm environment as that's all I have access to... This commit was SVN r25275.
2011-10-13 00:59:25 +04:00
AC_ARG_WITH([pmi],
[AC_HELP_STRING([--with-pmi],
[Build PMI support (default: no)])],
[], with_pmi=no)
Per the meeting on moving the BTLs to OPAL, move the ORTE database "db" framework to OPAL so the relocated BTLs can access it. Because the data is indexed by process, this requires that we define a new "opal_identifier_t" that corresponds to the orte_process_name_t struct. In order to support multiple run-times, this is defined in opal/mca/db/db_types.h as a uint64_t without identifying the meaning of any part of that data. A few changes were required to support this move: 1. the PMI component used to identify rte-related data (e.g., host name, bind level) and package them as a unit to reduce the number of PMI keys. This code was moved up to the ORTE layer as the OPAL layer has no understanding of these concepts. In addition, the component locally stored data based on process jobid/vpid - this could no longer be supported (see below for the solution). 2. the hash component was updated to use the new opal_identifier_t instead of orte_process_name_t as its index for storing data in the hash tables. Previously, we did a hash on the vpid and stored the data in a 32-bit hash table. In the revised system, we don't see a separate "vpid" field - we only have a 64-bit opaque value. The orte_process_name_t hash turned out to do nothing useful, so we now store the data in a 64-bit hash table. Preliminary tests didn't show any identifiable change in behavior or performance, but we'll have to see if a move back to the 32-bit table is required at some later time. 3. the db framework was a "select one" system. However, since the PMI component could no longer use its internal storage system, the framework has now been changed to a "select many" mode of operation. This allows the hash component to handle all internal storage, while the PMI component only handles pushing/pulling things from the PMI system. This was something we had planned for some time - when fetching data, we first check internal storage to see if we already have it, and then automatically go to the global system to look for it if we don't. Accordingly, the framework was provided with a custom query function used during "select" that lets you seperately specify the "store" and "fetch" ordering. 4. the ORTE grpcomm and ess/pmi components, and the nidmap code, were updated to work with the new db framework and to specify internal/global storage options. No changes were made to the MPI layer, except for modifying the ORTE component of the OMPI/rte framework to support the new db framework. This commit was SVN r28112.
2013-02-26 21:50:04 +04:00
opal_enable_pmi=0
Add 'generic' support for PMI2 (previously, we checked for PMI2 only on Cray systems.) If your resource manager (e.g. SLURM) has support for PMI2, then the --with-pmi configure flag will enable its usage. If you don't have PMI2, then you will fallback to regular old PMI1. This patch was submitted by Ralph Castain and reviewed and pushed by Josh Ladd. This should be added to cmr:v1.7:reviewer=jladd This commit was SVN r28666.
2013-06-21 19:28:14 +04:00
opal_pmi_rpath=
Cleanup a few things in the revised PMI configury - we know slurm has both pmi and pmi2 libs, so just auto-detect the presence of them if the user directed us to build with pmi support. Also cleanup some changed names in the alps code This commit was SVN r28670.
2013-06-24 06:41:40 +04:00
opal_have_pmi2=0
There is no convention out there as to how OEMs handle PMI2 functions. Some put them in their own -lpmi2 library, and some don't. Some have split the PMI2 definitions into a pmi2.h and keep the PMI-1 definitions in a separate pmi.h, and some don't. Try to handle cases more generally so at least Slurm and Cray can co-exist in peace. This commit was SVN r28672.
2013-06-26 04:43:26 +04:00
opal_have_pmi1=0
Complete the PMI support. Generalize PMI operations to support both slurm and non-slurm environments. Correct some configuration issues - we really only want the PMI integration at the individual component level. Ensure that the pmi grpcomm component doesn't get selected when launching via mpirun by setting its priority below the bad component. Only verified in a slurm environment as that's all I have access to... This commit was SVN r25275.
2011-10-13 00:59:25 +04:00
# save flags
Per the meeting on moving the BTLs to OPAL, move the ORTE database "db" framework to OPAL so the relocated BTLs can access it. Because the data is indexed by process, this requires that we define a new "opal_identifier_t" that corresponds to the orte_process_name_t struct. In order to support multiple run-times, this is defined in opal/mca/db/db_types.h as a uint64_t without identifying the meaning of any part of that data. A few changes were required to support this move: 1. the PMI component used to identify rte-related data (e.g., host name, bind level) and package them as a unit to reduce the number of PMI keys. This code was moved up to the ORTE layer as the OPAL layer has no understanding of these concepts. In addition, the component locally stored data based on process jobid/vpid - this could no longer be supported (see below for the solution). 2. the hash component was updated to use the new opal_identifier_t instead of orte_process_name_t as its index for storing data in the hash tables. Previously, we did a hash on the vpid and stored the data in a 32-bit hash table. In the revised system, we don't see a separate "vpid" field - we only have a 64-bit opaque value. The orte_process_name_t hash turned out to do nothing useful, so we now store the data in a 64-bit hash table. Preliminary tests didn't show any identifiable change in behavior or performance, but we'll have to see if a move back to the 32-bit table is required at some later time. 3. the db framework was a "select one" system. However, since the PMI component could no longer use its internal storage system, the framework has now been changed to a "select many" mode of operation. This allows the hash component to handle all internal storage, while the PMI component only handles pushing/pulling things from the PMI system. This was something we had planned for some time - when fetching data, we first check internal storage to see if we already have it, and then automatically go to the global system to look for it if we don't. Accordingly, the framework was provided with a custom query function used during "select" that lets you seperately specify the "store" and "fetch" ordering. 4. the ORTE grpcomm and ess/pmi components, and the nidmap code, were updated to work with the new db framework and to specify internal/global storage options. No changes were made to the MPI layer, except for modifying the ORTE component of the OMPI/rte framework to support the new db framework. This commit was SVN r28112.
2013-02-26 21:50:04 +04:00
opal_check_pmi_$1_save_CPPFLAGS="$CPPFLAGS"
opal_check_pmi_$1_save_LDFLAGS="$LDFLAGS"
opal_check_pmi_$1_save_LIBS="$LIBS"
Complete the PMI support. Generalize PMI operations to support both slurm and non-slurm environments. Correct some configuration issues - we really only want the PMI integration at the individual component level. Ensure that the pmi grpcomm component doesn't get selected when launching via mpirun by setting its priority below the bad component. Only verified in a slurm environment as that's all I have access to... This commit was SVN r25275.
2011-10-13 00:59:25 +04:00
# set defaults
Per the meeting on moving the BTLs to OPAL, move the ORTE database "db" framework to OPAL so the relocated BTLs can access it. Because the data is indexed by process, this requires that we define a new "opal_identifier_t" that corresponds to the orte_process_name_t struct. In order to support multiple run-times, this is defined in opal/mca/db/db_types.h as a uint64_t without identifying the meaning of any part of that data. A few changes were required to support this move: 1. the PMI component used to identify rte-related data (e.g., host name, bind level) and package them as a unit to reduce the number of PMI keys. This code was moved up to the ORTE layer as the OPAL layer has no understanding of these concepts. In addition, the component locally stored data based on process jobid/vpid - this could no longer be supported (see below for the solution). 2. the hash component was updated to use the new opal_identifier_t instead of orte_process_name_t as its index for storing data in the hash tables. Previously, we did a hash on the vpid and stored the data in a 32-bit hash table. In the revised system, we don't see a separate "vpid" field - we only have a 64-bit opaque value. The orte_process_name_t hash turned out to do nothing useful, so we now store the data in a 64-bit hash table. Preliminary tests didn't show any identifiable change in behavior or performance, but we'll have to see if a move back to the 32-bit table is required at some later time. 3. the db framework was a "select one" system. However, since the PMI component could no longer use its internal storage system, the framework has now been changed to a "select many" mode of operation. This allows the hash component to handle all internal storage, while the PMI component only handles pushing/pulling things from the PMI system. This was something we had planned for some time - when fetching data, we first check internal storage to see if we already have it, and then automatically go to the global system to look for it if we don't. Accordingly, the framework was provided with a custom query function used during "select" that lets you seperately specify the "store" and "fetch" ordering. 4. the ORTE grpcomm and ess/pmi components, and the nidmap code, were updated to work with the new db framework and to specify internal/global storage options. No changes were made to the MPI layer, except for modifying the ORTE component of the OMPI/rte framework to support the new db framework. This commit was SVN r28112.
2013-02-26 21:50:04 +04:00
opal_check_pmi_$1_LDFLAGS=
opal_check_pmi_$1_CPPFLAGS=
opal_check_pmi_$1_LIBS=
Complete the PMI support. Generalize PMI operations to support both slurm and non-slurm environments. Correct some configuration issues - we really only want the PMI integration at the individual component level. Ensure that the pmi grpcomm component doesn't get selected when launching via mpirun by setting its priority below the bad component. Only verified in a slurm environment as that's all I have access to... This commit was SVN r25275.
2011-10-13 00:59:25 +04:00
AC_MSG_CHECKING([if user requested PMI support])
Cleanup a few things in the revised PMI configury - we know slurm has both pmi and pmi2 libs, so just auto-detect the presence of them if the user directed us to build with pmi support. Also cleanup some changed names in the alps code This commit was SVN r28670.
2013-06-24 06:41:40 +04:00
opal_have_pmi_support=no
Complete the PMI support. Generalize PMI operations to support both slurm and non-slurm environments. Correct some configuration issues - we really only want the PMI integration at the individual component level. Ensure that the pmi grpcomm component doesn't get selected when launching via mpirun by setting its priority below the bad component. Only verified in a slurm environment as that's all I have access to... This commit was SVN r25275.
2011-10-13 00:59:25 +04:00
AS_IF([test "$with_pmi" = "no"],
[AC_MSG_RESULT([no])
update ess alps configuration file to include CNOS and PMI checks. some of the features committed here aren't being used, but they will be. also update orte_check_pmi.m4 to include missing call to action-if-not-found if --with-pmi is not specified or is disabled. This commit was SVN r25422.
2011-11-03 06:14:47 +04:00
$3],
Complete the PMI support. Generalize PMI operations to support both slurm and non-slurm environments. Correct some configuration issues - we really only want the PMI integration at the individual component level. Ensure that the pmi grpcomm component doesn't get selected when launching via mpirun by setting its priority below the bad component. Only verified in a slurm environment as that's all I have access to... This commit was SVN r25275.
2011-10-13 00:59:25 +04:00
[AC_MSG_RESULT([yes])
Cleanup a few things in the revised PMI configury - we know slurm has both pmi and pmi2 libs, so just auto-detect the presence of them if the user directed us to build with pmi support. Also cleanup some changed names in the alps code This commit was SVN r28670.
2013-06-24 06:41:40 +04:00
AC_MSG_CHECKING([if PMI or PMI2 headers installed])
Complete the PMI support. Generalize PMI operations to support both slurm and non-slurm environments. Correct some configuration issues - we really only want the PMI integration at the individual component level. Ensure that the pmi grpcomm component doesn't get selected when launching via mpirun by setting its priority below the bad component. Only verified in a slurm environment as that's all I have access to... This commit was SVN r25275.
2011-10-13 00:59:25 +04:00
# cannot use OMPI_CHECK_PACKAGE as its backend header
# support appends "include" to the path, which won't
# work with slurm :-(
AS_IF([test ! -z "$with_pmi" -a "$with_pmi" != "yes"],
[AS_IF([test -d "$with_pmi/lib64"],
Per the meeting on moving the BTLs to OPAL, move the ORTE database "db" framework to OPAL so the relocated BTLs can access it. Because the data is indexed by process, this requires that we define a new "opal_identifier_t" that corresponds to the orte_process_name_t struct. In order to support multiple run-times, this is defined in opal/mca/db/db_types.h as a uint64_t without identifying the meaning of any part of that data. A few changes were required to support this move: 1. the PMI component used to identify rte-related data (e.g., host name, bind level) and package them as a unit to reduce the number of PMI keys. This code was moved up to the ORTE layer as the OPAL layer has no understanding of these concepts. In addition, the component locally stored data based on process jobid/vpid - this could no longer be supported (see below for the solution). 2. the hash component was updated to use the new opal_identifier_t instead of orte_process_name_t as its index for storing data in the hash tables. Previously, we did a hash on the vpid and stored the data in a 32-bit hash table. In the revised system, we don't see a separate "vpid" field - we only have a 64-bit opaque value. The orte_process_name_t hash turned out to do nothing useful, so we now store the data in a 64-bit hash table. Preliminary tests didn't show any identifiable change in behavior or performance, but we'll have to see if a move back to the 32-bit table is required at some later time. 3. the db framework was a "select one" system. However, since the PMI component could no longer use its internal storage system, the framework has now been changed to a "select many" mode of operation. This allows the hash component to handle all internal storage, while the PMI component only handles pushing/pulling things from the PMI system. This was something we had planned for some time - when fetching data, we first check internal storage to see if we already have it, and then automatically go to the global system to look for it if we don't. Accordingly, the framework was provided with a custom query function used during "select" that lets you seperately specify the "store" and "fetch" ordering. 4. the ORTE grpcomm and ess/pmi components, and the nidmap code, were updated to work with the new db framework and to specify internal/global storage options. No changes were made to the MPI layer, except for modifying the ORTE component of the OMPI/rte framework to support the new db framework. This commit was SVN r28112.
2013-02-26 21:50:04 +04:00
[opal_check_pmi_$1_LDFLAGS="-L$with_pmi/lib64"
Add 'generic' support for PMI2 (previously, we checked for PMI2 only on Cray systems.) If your resource manager (e.g. SLURM) has support for PMI2, then the --with-pmi configure flag will enable its usage. If you don't have PMI2, then you will fallback to regular old PMI1. This patch was submitted by Ralph Castain and reviewed and pushed by Josh Ladd. This should be added to cmr:v1.7:reviewer=jladd This commit was SVN r28666.
2013-06-21 19:28:14 +04:00
opal_pmi_rpath="$with_pmi/lib64"],
Per the meeting on moving the BTLs to OPAL, move the ORTE database "db" framework to OPAL so the relocated BTLs can access it. Because the data is indexed by process, this requires that we define a new "opal_identifier_t" that corresponds to the orte_process_name_t struct. In order to support multiple run-times, this is defined in opal/mca/db/db_types.h as a uint64_t without identifying the meaning of any part of that data. A few changes were required to support this move: 1. the PMI component used to identify rte-related data (e.g., host name, bind level) and package them as a unit to reduce the number of PMI keys. This code was moved up to the ORTE layer as the OPAL layer has no understanding of these concepts. In addition, the component locally stored data based on process jobid/vpid - this could no longer be supported (see below for the solution). 2. the hash component was updated to use the new opal_identifier_t instead of orte_process_name_t as its index for storing data in the hash tables. Previously, we did a hash on the vpid and stored the data in a 32-bit hash table. In the revised system, we don't see a separate "vpid" field - we only have a 64-bit opaque value. The orte_process_name_t hash turned out to do nothing useful, so we now store the data in a 64-bit hash table. Preliminary tests didn't show any identifiable change in behavior or performance, but we'll have to see if a move back to the 32-bit table is required at some later time. 3. the db framework was a "select one" system. However, since the PMI component could no longer use its internal storage system, the framework has now been changed to a "select many" mode of operation. This allows the hash component to handle all internal storage, while the PMI component only handles pushing/pulling things from the PMI system. This was something we had planned for some time - when fetching data, we first check internal storage to see if we already have it, and then automatically go to the global system to look for it if we don't. Accordingly, the framework was provided with a custom query function used during "select" that lets you seperately specify the "store" and "fetch" ordering. 4. the ORTE grpcomm and ess/pmi components, and the nidmap code, were updated to work with the new db framework and to specify internal/global storage options. No changes were made to the MPI layer, except for modifying the ORTE component of the OMPI/rte framework to support the new db framework. This commit was SVN r28112.
2013-02-26 21:50:04 +04:00
[opal_check_pmi_$1_LDFLAGS="-L$with_pmi/lib"
Add 'generic' support for PMI2 (previously, we checked for PMI2 only on Cray systems.) If your resource manager (e.g. SLURM) has support for PMI2, then the --with-pmi configure flag will enable its usage. If you don't have PMI2, then you will fallback to regular old PMI1. This patch was submitted by Ralph Castain and reviewed and pushed by Josh Ladd. This should be added to cmr:v1.7:reviewer=jladd This commit was SVN r28666.
2013-06-21 19:28:14 +04:00
opal_pmi_rpath="$with_pmi/lib"])
Cleanup a few things in the revised PMI configury - we know slurm has both pmi and pmi2 libs, so just auto-detect the presence of them if the user directed us to build with pmi support. Also cleanup some changed names in the alps code This commit was SVN r28670.
2013-06-24 06:41:40 +04:00
# look for required headers - both pmi.h AND/OR pmi2.h
# may be present
Add 'generic' support for PMI2 (previously, we checked for PMI2 only on Cray systems.) If your resource manager (e.g. SLURM) has support for PMI2, then the --with-pmi configure flag will enable its usage. If you don't have PMI2, then you will fallback to regular old PMI1. This patch was submitted by Ralph Castain and reviewed and pushed by Josh Ladd. This should be added to cmr:v1.7:reviewer=jladd This commit was SVN r28666.
2013-06-21 19:28:14 +04:00
AS_IF([test -f "$with_pmi/include/pmi2.h" -o -f "$with_pmi/include/pmi.h"],
[opal_check_pmi_$1_CPPFLAGS="-I$with_pmi/include"
AS_IF([test -f "$with_pmi/include/pmi2.h"],
There is no convention out there as to how OEMs handle PMI2 functions. Some put them in their own -lpmi2 library, and some don't. Some have split the PMI2 definitions into a pmi2.h and keep the PMI-1 definitions in a separate pmi.h, and some don't. Try to handle cases more generally so at least Slurm and Cray can co-exist in peace. This commit was SVN r28672.
2013-06-26 04:43:26 +04:00
[AC_MSG_RESULT([PMI2 header found])],
[AC_MSG_RESULT([PMI header found])])],
Add 'generic' support for PMI2 (previously, we checked for PMI2 only on Cray systems.) If your resource manager (e.g. SLURM) has support for PMI2, then the --with-pmi configure flag will enable its usage. If you don't have PMI2, then you will fallback to regular old PMI1. This patch was submitted by Ralph Castain and reviewed and pushed by Josh Ladd. This should be added to cmr:v1.7:reviewer=jladd This commit was SVN r28666.
2013-06-21 19:28:14 +04:00
[AS_IF([test -f "$with_pmi/include/slurm/pmi2.h" -o -f "$with_pmi/include/slurm/pmi.h"],
[opal_check_pmi_$1_CPPFLAGS="-I$with_pmi/include/slurm"
AS_IF([test -f "$with_pmi/include/slurm/pmi2.h"],
There is no convention out there as to how OEMs handle PMI2 functions. Some put them in their own -lpmi2 library, and some don't. Some have split the PMI2 definitions into a pmi2.h and keep the PMI-1 definitions in a separate pmi.h, and some don't. Try to handle cases more generally so at least Slurm and Cray can co-exist in peace. This commit was SVN r28672.
2013-06-26 04:43:26 +04:00
[AC_MSG_RESULT([Slurm PMI2 headers found])],
[AC_MSG_RESULT([Slurm PMI headers found])])],
Update the pmi detection to report a more meaningful error message when with-pmi is specified but the pmi.h file cannot be found This commit was SVN r27725.
2012-12-27 08:57:30 +04:00
[AC_MSG_RESULT([not found])
Add 'generic' support for PMI2 (previously, we checked for PMI2 only on Cray systems.) If your resource manager (e.g. SLURM) has support for PMI2, then the --with-pmi configure flag will enable its usage. If you don't have PMI2, then you will fallback to regular old PMI1. This patch was submitted by Ralph Castain and reviewed and pushed by Josh Ladd. This should be added to cmr:v1.7:reviewer=jladd This commit was SVN r28666.
2013-06-21 19:28:14 +04:00
AC_MSG_WARN([PMI support requested (via --with-pmi) but neither pmi.h])
AC_MSG_WARN([nor pmi2.h were found under locations:])
Improve the error message a bit This commit was SVN r27726.
2012-12-28 07:02:07 +04:00
AC_MSG_WARN([ $with_pmi/include])
AC_MSG_WARN([ $with_pmi/include/slurm])
AC_MSG_WARN([Specified path: $with_pmi])
Update the pmi detection to report a more meaningful error message when with-pmi is specified but the pmi.h file cannot be found This commit was SVN r27725.
2012-12-27 08:57:30 +04:00
AC_MSG_ERROR([Aborting])
Cleanup a few things in the revised PMI configury - we know slurm has both pmi and pmi2 libs, so just auto-detect the presence of them if the user directed us to build with pmi support. Also cleanup some changed names in the alps code This commit was SVN r28670.
2013-06-24 06:41:40 +04:00
$3])])],
[AS_IF([test -f "/usr/include/slurm/pmi2.h" -o -f "/usr/include/slurm/pmi.h"],
[opal_check_pmi_$1_CPPFLAGS="-I/usr/include/slurm"])
AC_MSG_RESULT([in default locations])])
Cray's PMI implementation is quite different from slurm's - they extended PMI-1 by adding some, but not all, of the PMI-2 APIs. So you can't just switch to using PMI-2 functions as it isn't a complete implementation. Instead, you have to selectively figure out which ones they have in PMI-2, and use any missing ones from PMI-1. What fun. Modify the configure logic and the PMI components to accommodate Cray's approach. Refactor the PMI error reporting code so it resides in only one place. Cray actually decided -not- to define the PMI-2 error codes, so we have to use the PMI-1 codes instead. More fun. This commit was SVN r25348.
2011-10-21 08:54:38 +04:00
Cleanup a few things in the revised PMI configury - we know slurm has both pmi and pmi2 libs, so just auto-detect the presence of them if the user directed us to build with pmi support. Also cleanup some changed names in the alps code This commit was SVN r28670.
2013-06-24 06:41:40 +04:00
# setup to check libraries
Per the meeting on moving the BTLs to OPAL, move the ORTE database "db" framework to OPAL so the relocated BTLs can access it. Because the data is indexed by process, this requires that we define a new "opal_identifier_t" that corresponds to the orte_process_name_t struct. In order to support multiple run-times, this is defined in opal/mca/db/db_types.h as a uint64_t without identifying the meaning of any part of that data. A few changes were required to support this move: 1. the PMI component used to identify rte-related data (e.g., host name, bind level) and package them as a unit to reduce the number of PMI keys. This code was moved up to the ORTE layer as the OPAL layer has no understanding of these concepts. In addition, the component locally stored data based on process jobid/vpid - this could no longer be supported (see below for the solution). 2. the hash component was updated to use the new opal_identifier_t instead of orte_process_name_t as its index for storing data in the hash tables. Previously, we did a hash on the vpid and stored the data in a 32-bit hash table. In the revised system, we don't see a separate "vpid" field - we only have a 64-bit opaque value. The orte_process_name_t hash turned out to do nothing useful, so we now store the data in a 64-bit hash table. Preliminary tests didn't show any identifiable change in behavior or performance, but we'll have to see if a move back to the 32-bit table is required at some later time. 3. the db framework was a "select one" system. However, since the PMI component could no longer use its internal storage system, the framework has now been changed to a "select many" mode of operation. This allows the hash component to handle all internal storage, while the PMI component only handles pushing/pulling things from the PMI system. This was something we had planned for some time - when fetching data, we first check internal storage to see if we already have it, and then automatically go to the global system to look for it if we don't. Accordingly, the framework was provided with a custom query function used during "select" that lets you seperately specify the "store" and "fetch" ordering. 4. the ORTE grpcomm and ess/pmi components, and the nidmap code, were updated to work with the new db framework and to specify internal/global storage options. No changes were made to the MPI layer, except for modifying the ORTE component of the OMPI/rte framework to support the new db framework. This commit was SVN r28112.
2013-02-26 21:50:04 +04:00
LDFLAGS="$LDFLAGS $opal_check_pmi_$1_LDFLAGS"
CPPFLAGS="$CPPFLAGS $opal_check_pmi_$1_CPPFLAGS"
Cleanup a few things in the revised PMI configury - we know slurm has both pmi and pmi2 libs, so just auto-detect the presence of them if the user directed us to build with pmi support. Also cleanup some changed names in the alps code This commit was SVN r28670.
2013-06-24 06:41:40 +04:00
# reset the included libs so we only link in the
# ones we successfully check
Ensure we get both -lpmi and -lpmi2 when the libs are separate This commit was SVN r28795.
2013-07-16 18:57:18 +04:00
opal_check_pmi_$1_LIBS=
Cleanup a few things in the revised PMI configury - we know slurm has both pmi and pmi2 libs, so just auto-detect the presence of them if the user directed us to build with pmi support. Also cleanup some changed names in the alps code This commit was SVN r28670.
2013-06-24 06:41:40 +04:00
# check the PMI libs - both -lpmi and -lpmi2 may
# be present. If both are present, then we need
# to link against both
There is no convention out there as to how OEMs handle PMI2 functions. Some put them in their own -lpmi2 library, and some don't. Some have split the PMI2 definitions into a pmi2.h and keep the PMI-1 definitions in a separate pmi.h, and some don't. Try to handle cases more generally so at least Slurm and Cray can co-exist in peace. This commit was SVN r28672.
2013-06-26 04:43:26 +04:00
LIBS="$opal_check_pmi_$1_save_LIBS -lpmi2"
Cleanup a few things in the revised PMI configury - we know slurm has both pmi and pmi2 libs, so just auto-detect the presence of them if the user directed us to build with pmi support. Also cleanup some changed names in the alps code This commit was SVN r28670.
2013-06-24 06:41:40 +04:00
AC_CHECK_LIB([pmi2], [PMI2_Init],
[opal_have_pmi_support=yes
opal_have_pmi2=1
Ensure we get both -lpmi and -lpmi2 when the libs are separate This commit was SVN r28795.
2013-07-16 18:57:18 +04:00
opal_check_pmi_$1_LIBS="$opal_check_pmi_$1_LIBS -lpmi2"])
There is no convention out there as to how OEMs handle PMI2 functions. Some put them in their own -lpmi2 library, and some don't. Some have split the PMI2 definitions into a pmi2.h and keep the PMI-1 definitions in a separate pmi.h, and some don't. Try to handle cases more generally so at least Slurm and Cray can co-exist in peace. This commit was SVN r28672.
2013-06-26 04:43:26 +04:00
# if the pmi2 functions aren't in -lpmi2, they might
# be in -lpmi. Nobody follows a convention here, so
# all we can do is check both
AS_IF([test "$opal_have_pmi2" = "0"],
[LIBS="$opal_check_pmi_$1_save_LIBS -lpmi"
AC_CHECK_LIB([pmi], [PMI2_Init],
[opal_have_pmi_support=yes
opal_have_pmi2=1
opal_have_pmi1=1
Ensure we get both -lpmi and -lpmi2 when the libs are separate This commit was SVN r28795.
2013-07-16 18:57:18 +04:00
opal_check_pmi_$1_LIBS="$opal_check_pmi_$1_LIBS -lpmi"])])
There is no convention out there as to how OEMs handle PMI2 functions. Some put them in their own -lpmi2 library, and some don't. Some have split the PMI2 definitions into a pmi2.h and keep the PMI-1 definitions in a separate pmi.h, and some don't. Try to handle cases more generally so at least Slurm and Cray can co-exist in peace. This commit was SVN r28672.
2013-06-26 04:43:26 +04:00
# if we haven't already added -lpmi, look for the pmi1 functions
AS_IF([test "$opal_have_pmi1" = "0"],
[LIBS="$opal_check_pmi_$1_save_LIBS -lpmi"
AC_CHECK_LIB([pmi], [PMI_Init],
[opal_have_pmi_support=yes
Ensure we get both -lpmi and -lpmi2 when the libs are separate This commit was SVN r28795.
2013-07-16 18:57:18 +04:00
opal_check_pmi_$1_LIBS="$opal_check_pmi_$1_LIBS -lpmi"])])
Complete the PMI support. Generalize PMI operations to support both slurm and non-slurm environments. Correct some configuration issues - we really only want the PMI integration at the individual component level. Ensure that the pmi grpcomm component doesn't get selected when launching via mpirun by setting its priority below the bad component. Only verified in a slurm environment as that's all I have access to... This commit was SVN r25275.
2011-10-13 00:59:25 +04:00
There is no convention out there as to how OEMs handle PMI2 functions. Some put them in their own -lpmi2 library, and some don't. Some have split the PMI2 definitions into a pmi2.h and keep the PMI-1 definitions in a separate pmi.h, and some don't. Try to handle cases more generally so at least Slurm and Cray can co-exist in peace. This commit was SVN r28672.
2013-06-26 04:43:26 +04:00
AC_MSG_CHECKING([PMI2 and/or PMI support enabled])
Per the meeting on moving the BTLs to OPAL, move the ORTE database "db" framework to OPAL so the relocated BTLs can access it. Because the data is indexed by process, this requires that we define a new "opal_identifier_t" that corresponds to the orte_process_name_t struct. In order to support multiple run-times, this is defined in opal/mca/db/db_types.h as a uint64_t without identifying the meaning of any part of that data. A few changes were required to support this move: 1. the PMI component used to identify rte-related data (e.g., host name, bind level) and package them as a unit to reduce the number of PMI keys. This code was moved up to the ORTE layer as the OPAL layer has no understanding of these concepts. In addition, the component locally stored data based on process jobid/vpid - this could no longer be supported (see below for the solution). 2. the hash component was updated to use the new opal_identifier_t instead of orte_process_name_t as its index for storing data in the hash tables. Previously, we did a hash on the vpid and stored the data in a 32-bit hash table. In the revised system, we don't see a separate "vpid" field - we only have a 64-bit opaque value. The orte_process_name_t hash turned out to do nothing useful, so we now store the data in a 64-bit hash table. Preliminary tests didn't show any identifiable change in behavior or performance, but we'll have to see if a move back to the 32-bit table is required at some later time. 3. the db framework was a "select one" system. However, since the PMI component could no longer use its internal storage system, the framework has now been changed to a "select many" mode of operation. This allows the hash component to handle all internal storage, while the PMI component only handles pushing/pulling things from the PMI system. This was something we had planned for some time - when fetching data, we first check internal storage to see if we already have it, and then automatically go to the global system to look for it if we don't. Accordingly, the framework was provided with a custom query function used during "select" that lets you seperately specify the "store" and "fetch" ordering. 4. the ORTE grpcomm and ess/pmi components, and the nidmap code, were updated to work with the new db framework and to specify internal/global storage options. No changes were made to the MPI layer, except for modifying the ORTE component of the OMPI/rte framework to support the new db framework. This commit was SVN r28112.
2013-02-26 21:50:04 +04:00
AS_IF([test "$opal_have_pmi_support" = "yes"],
Complete the PMI support. Generalize PMI operations to support both slurm and non-slurm environments. Correct some configuration issues - we really only want the PMI integration at the individual component level. Ensure that the pmi grpcomm component doesn't get selected when launching via mpirun by setting its priority below the bad component. Only verified in a slurm environment as that's all I have access to... This commit was SVN r25275.
2011-10-13 00:59:25 +04:00
[AC_MSG_RESULT([yes])
Per the meeting on moving the BTLs to OPAL, move the ORTE database "db" framework to OPAL so the relocated BTLs can access it. Because the data is indexed by process, this requires that we define a new "opal_identifier_t" that corresponds to the orte_process_name_t struct. In order to support multiple run-times, this is defined in opal/mca/db/db_types.h as a uint64_t without identifying the meaning of any part of that data. A few changes were required to support this move: 1. the PMI component used to identify rte-related data (e.g., host name, bind level) and package them as a unit to reduce the number of PMI keys. This code was moved up to the ORTE layer as the OPAL layer has no understanding of these concepts. In addition, the component locally stored data based on process jobid/vpid - this could no longer be supported (see below for the solution). 2. the hash component was updated to use the new opal_identifier_t instead of orte_process_name_t as its index for storing data in the hash tables. Previously, we did a hash on the vpid and stored the data in a 32-bit hash table. In the revised system, we don't see a separate "vpid" field - we only have a 64-bit opaque value. The orte_process_name_t hash turned out to do nothing useful, so we now store the data in a 64-bit hash table. Preliminary tests didn't show any identifiable change in behavior or performance, but we'll have to see if a move back to the 32-bit table is required at some later time. 3. the db framework was a "select one" system. However, since the PMI component could no longer use its internal storage system, the framework has now been changed to a "select many" mode of operation. This allows the hash component to handle all internal storage, while the PMI component only handles pushing/pulling things from the PMI system. This was something we had planned for some time - when fetching data, we first check internal storage to see if we already have it, and then automatically go to the global system to look for it if we don't. Accordingly, the framework was provided with a custom query function used during "select" that lets you seperately specify the "store" and "fetch" ordering. 4. the ORTE grpcomm and ess/pmi components, and the nidmap code, were updated to work with the new db framework and to specify internal/global storage options. No changes were made to the MPI layer, except for modifying the ORTE component of the OMPI/rte framework to support the new db framework. This commit was SVN r28112.
2013-02-26 21:50:04 +04:00
opal_enable_pmi=1
$1_LDFLAGS="$opal_check_pmi_$1_LDFLAGS"
$1_CPPFLAGS="$opal_check_pmi_$1_CPPFLAGS"
Cleanup a few things in the revised PMI configury - we know slurm has both pmi and pmi2 libs, so just auto-detect the presence of them if the user directed us to build with pmi support. Also cleanup some changed names in the alps code This commit was SVN r28670.
2013-06-24 06:41:40 +04:00
$1_LIBS="$opal_check_pmi_$1_LIBS -Wl,-rpath=$opal_pmi_rpath"
Ensure we get both -lpmi and -lpmi2 when the libs are separate This commit was SVN r28795.
2013-07-16 18:57:18 +04:00
AC_MSG_CHECKING([final added libraries])
AC_MSG_RESULT([$opal_check_pmi_$1_LIBS])
Complete the PMI support. Generalize PMI operations to support both slurm and non-slurm environments. Correct some configuration issues - we really only want the PMI integration at the individual component level. Ensure that the pmi grpcomm component doesn't get selected when launching via mpirun by setting its priority below the bad component. Only verified in a slurm environment as that's all I have access to... This commit was SVN r25275.
2011-10-13 00:59:25 +04:00
$2],
[AC_MSG_RESULT([no])
AC_MSG_WARN([PMI support requested (via --with-pmi) but not found.])
AC_MSG_ERROR([Aborting.])
Cray's PMI implementation is quite different from slurm's - they extended PMI-1 by adding some, but not all, of the PMI-2 APIs. So you can't just switch to using PMI-2 functions as it isn't a complete implementation. Instead, you have to selectively figure out which ones they have in PMI-2, and use any missing ones from PMI-1. What fun. Modify the configure logic and the PMI components to accommodate Cray's approach. Refactor the PMI error reporting code so it resides in only one place. Cray actually decided -not- to define the PMI-2 error codes, so we have to use the PMI-1 codes instead. More fun. This commit was SVN r25348.
2011-10-21 08:54:38 +04:00
$3])
Add 'generic' support for PMI2 (previously, we checked for PMI2 only on Cray systems.) If your resource manager (e.g. SLURM) has support for PMI2, then the --with-pmi configure flag will enable its usage. If you don't have PMI2, then you will fallback to regular old PMI1. This patch was submitted by Ralph Castain and reviewed and pushed by Josh Ladd. This should be added to cmr:v1.7:reviewer=jladd This commit was SVN r28666.
2013-06-21 19:28:14 +04:00
])
Complete the PMI support. Generalize PMI operations to support both slurm and non-slurm environments. Correct some configuration issues - we really only want the PMI integration at the individual component level. Ensure that the pmi grpcomm component doesn't get selected when launching via mpirun by setting its priority below the bad component. Only verified in a slurm environment as that's all I have access to... This commit was SVN r25275.
2011-10-13 00:59:25 +04:00
Fixes trac:3486 - prevent opal_check_pmi from bleeding CPPFLAGS This commit was SVN r28940. The following Trac tickets were found above: Ticket 3486 --> https://svn.open-mpi.org/trac/ompi/ticket/3486
2013-07-24 07:53:23 +04:00
# restore flags
CPPFLAGS="$opal_check_pmi_$1_save_CPPFLAGS"
Per the meeting on moving the BTLs to OPAL, move the ORTE database "db" framework to OPAL so the relocated BTLs can access it. Because the data is indexed by process, this requires that we define a new "opal_identifier_t" that corresponds to the orte_process_name_t struct. In order to support multiple run-times, this is defined in opal/mca/db/db_types.h as a uint64_t without identifying the meaning of any part of that data. A few changes were required to support this move: 1. the PMI component used to identify rte-related data (e.g., host name, bind level) and package them as a unit to reduce the number of PMI keys. This code was moved up to the ORTE layer as the OPAL layer has no understanding of these concepts. In addition, the component locally stored data based on process jobid/vpid - this could no longer be supported (see below for the solution). 2. the hash component was updated to use the new opal_identifier_t instead of orte_process_name_t as its index for storing data in the hash tables. Previously, we did a hash on the vpid and stored the data in a 32-bit hash table. In the revised system, we don't see a separate "vpid" field - we only have a 64-bit opaque value. The orte_process_name_t hash turned out to do nothing useful, so we now store the data in a 64-bit hash table. Preliminary tests didn't show any identifiable change in behavior or performance, but we'll have to see if a move back to the 32-bit table is required at some later time. 3. the db framework was a "select one" system. However, since the PMI component could no longer use its internal storage system, the framework has now been changed to a "select many" mode of operation. This allows the hash component to handle all internal storage, while the PMI component only handles pushing/pulling things from the PMI system. This was something we had planned for some time - when fetching data, we first check internal storage to see if we already have it, and then automatically go to the global system to look for it if we don't. Accordingly, the framework was provided with a custom query function used during "select" that lets you seperately specify the "store" and "fetch" ordering. 4. the ORTE grpcomm and ess/pmi components, and the nidmap code, were updated to work with the new db framework and to specify internal/global storage options. No changes were made to the MPI layer, except for modifying the ORTE component of the OMPI/rte framework to support the new db framework. This commit was SVN r28112.
2013-02-26 21:50:04 +04:00
LDFLAGS="$opal_check_pmi_$1_save_LDFLAGS"
LIBS="$opal_check_pmi_$1_save_LIBS"
Complete the PMI support. Generalize PMI operations to support both slurm and non-slurm environments. Correct some configuration issues - we really only want the PMI integration at the individual component level. Ensure that the pmi grpcomm component doesn't get selected when launching via mpirun by setting its priority below the bad component. Only verified in a slurm environment as that's all I have access to... This commit was SVN r25275.
2011-10-13 00:59:25 +04:00
first cut at general pmi check This commit was SVN r25234.
2011-10-05 21:14:24 +04:00
AC_DEFINE_UNQUOTED([WANT_PMI_SUPPORT],
Per the meeting on moving the BTLs to OPAL, move the ORTE database "db" framework to OPAL so the relocated BTLs can access it. Because the data is indexed by process, this requires that we define a new "opal_identifier_t" that corresponds to the orte_process_name_t struct. In order to support multiple run-times, this is defined in opal/mca/db/db_types.h as a uint64_t without identifying the meaning of any part of that data. A few changes were required to support this move: 1. the PMI component used to identify rte-related data (e.g., host name, bind level) and package them as a unit to reduce the number of PMI keys. This code was moved up to the ORTE layer as the OPAL layer has no understanding of these concepts. In addition, the component locally stored data based on process jobid/vpid - this could no longer be supported (see below for the solution). 2. the hash component was updated to use the new opal_identifier_t instead of orte_process_name_t as its index for storing data in the hash tables. Previously, we did a hash on the vpid and stored the data in a 32-bit hash table. In the revised system, we don't see a separate "vpid" field - we only have a 64-bit opaque value. The orte_process_name_t hash turned out to do nothing useful, so we now store the data in a 64-bit hash table. Preliminary tests didn't show any identifiable change in behavior or performance, but we'll have to see if a move back to the 32-bit table is required at some later time. 3. the db framework was a "select one" system. However, since the PMI component could no longer use its internal storage system, the framework has now been changed to a "select many" mode of operation. This allows the hash component to handle all internal storage, while the PMI component only handles pushing/pulling things from the PMI system. This was something we had planned for some time - when fetching data, we first check internal storage to see if we already have it, and then automatically go to the global system to look for it if we don't. Accordingly, the framework was provided with a custom query function used during "select" that lets you seperately specify the "store" and "fetch" ordering. 4. the ORTE grpcomm and ess/pmi components, and the nidmap code, were updated to work with the new db framework and to specify internal/global storage options. No changes were made to the MPI layer, except for modifying the ORTE component of the OMPI/rte framework to support the new db framework. This commit was SVN r28112.
2013-02-26 21:50:04 +04:00
[$opal_enable_pmi],
first cut at general pmi check This commit was SVN r25234.
2011-10-05 21:14:24 +04:00
[Whether we want PMI support])
Add 'generic' support for PMI2 (previously, we checked for PMI2 only on Cray systems.) If your resource manager (e.g. SLURM) has support for PMI2, then the --with-pmi configure flag will enable its usage. If you don't have PMI2, then you will fallback to regular old PMI1. This patch was submitted by Ralph Castain and reviewed and pushed by Josh Ladd. This should be added to cmr:v1.7:reviewer=jladd This commit was SVN r28666.
2013-06-21 19:28:14 +04:00
AC_DEFINE_UNQUOTED([WANT_PMI2_SUPPORT],
Cleanup a few things in the revised PMI configury - we know slurm has both pmi and pmi2 libs, so just auto-detect the presence of them if the user directed us to build with pmi support. Also cleanup some changed names in the alps code This commit was SVN r28670.
2013-06-24 06:41:40 +04:00
[$opal_have_pmi2],
[Whether we have PMI2 support])
Per the meeting on moving the BTLs to OPAL, move the ORTE database "db" framework to OPAL so the relocated BTLs can access it. Because the data is indexed by process, this requires that we define a new "opal_identifier_t" that corresponds to the orte_process_name_t struct. In order to support multiple run-times, this is defined in opal/mca/db/db_types.h as a uint64_t without identifying the meaning of any part of that data. A few changes were required to support this move: 1. the PMI component used to identify rte-related data (e.g., host name, bind level) and package them as a unit to reduce the number of PMI keys. This code was moved up to the ORTE layer as the OPAL layer has no understanding of these concepts. In addition, the component locally stored data based on process jobid/vpid - this could no longer be supported (see below for the solution). 2. the hash component was updated to use the new opal_identifier_t instead of orte_process_name_t as its index for storing data in the hash tables. Previously, we did a hash on the vpid and stored the data in a 32-bit hash table. In the revised system, we don't see a separate "vpid" field - we only have a 64-bit opaque value. The orte_process_name_t hash turned out to do nothing useful, so we now store the data in a 64-bit hash table. Preliminary tests didn't show any identifiable change in behavior or performance, but we'll have to see if a move back to the 32-bit table is required at some later time. 3. the db framework was a "select one" system. However, since the PMI component could no longer use its internal storage system, the framework has now been changed to a "select many" mode of operation. This allows the hash component to handle all internal storage, while the PMI component only handles pushing/pulling things from the PMI system. This was something we had planned for some time - when fetching data, we first check internal storage to see if we already have it, and then automatically go to the global system to look for it if we don't. Accordingly, the framework was provided with a custom query function used during "select" that lets you seperately specify the "store" and "fetch" ordering. 4. the ORTE grpcomm and ess/pmi components, and the nidmap code, were updated to work with the new db framework and to specify internal/global storage options. No changes were made to the MPI layer, except for modifying the ORTE component of the OMPI/rte framework to support the new db framework. This commit was SVN r28112.
2013-02-26 21:50:04 +04:00
AM_CONDITIONAL(WANT_PMI_SUPPORT, [test "$opal_enable_pmi" = 1])
first cut at general pmi check This commit was SVN r25234.
2011-10-05 21:14:24 +04:00
])
Копировать постоянную ссылку