2013-07-20 02:13:58 +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.
|
2014-07-30 06:11:04 +04:00
|
|
|
# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
2013-07-20 02:13:58 +04:00
|
|
|
# University of Stuttgart. All rights reserved.
|
|
|
|
# Copyright (c) 2004-2005 The Regents of the University of California.
|
|
|
|
# All rights reserved.
|
|
|
|
# Copyright (c) 2006 Sandia National Laboratories. All rights
|
|
|
|
# reserved.
|
2014-01-10 02:17:01 +04:00
|
|
|
# Copyright (c) 2010-2014 Cisco Systems, Inc. All rights reserved.
|
2013-07-20 02:13:58 +04:00
|
|
|
# $COPYRIGHT$
|
2014-07-30 06:11:04 +04:00
|
|
|
#
|
2013-07-20 02:13:58 +04:00
|
|
|
# Additional copyrights may follow
|
2014-07-30 06:11:04 +04:00
|
|
|
#
|
2013-07-20 02:13:58 +04:00
|
|
|
# $HEADER$
|
|
|
|
#
|
|
|
|
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 04:47:28 +04:00
|
|
|
# OPAL_CHECK_LIBNL3(prefix, [action-if-found], [action-if-not-found])
|
2014-02-26 11:47:50 +04:00
|
|
|
# --------------------------------------------------------
|
|
|
|
# check if libnl3 support can be found. sets prefix_{CPPFLAGS,
|
|
|
|
# LDFLAGS, LIBS} as needed and runs action-if-found if there is
|
|
|
|
# support, otherwise executes action-if-not-found
|
|
|
|
#
|
|
|
|
# libnl3 changed its default header location as of v3.2 (released ca. September
|
|
|
|
# 2011). It was previously "${prefix}/include/netlink/...". It now is
|
|
|
|
# "${prefix}/libnl3/include/netlink/...". The logic below only supports
|
|
|
|
# >=v3.2, under the assumption that it is not widely deployed.
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 04:47:28 +04:00
|
|
|
AC_DEFUN([OPAL_CHECK_LIBNL3],[
|
2014-02-26 11:47:50 +04:00
|
|
|
AC_ARG_WITH([libnl3],
|
|
|
|
[AC_HELP_STRING([--with-libnl3(=DIR)],
|
|
|
|
[Build libnl3 support])])
|
2014-05-01 19:38:07 +04:00
|
|
|
OPAL_CHECK_WITHDIR([libnl3], [$with_libnl3], [include/libnl3/netlink/netlink.h])
|
2014-02-26 11:47:50 +04:00
|
|
|
AC_ARG_WITH([libnl3-libdir],
|
|
|
|
[AC_HELP_STRING([--with-libnl3-libdir=DIR],
|
|
|
|
[Search for libnl3 libraries in DIR])])
|
2014-05-01 19:38:07 +04:00
|
|
|
OPAL_CHECK_WITHDIR([libnl3-libdir], [$with_libnl3_libdir], [libnl-3.*])
|
2014-02-26 11:47:50 +04:00
|
|
|
|
|
|
|
ompi_check_libnl3_$1_save_CPPFLAGS="$CPPFLAGS"
|
|
|
|
ompi_check_libnl3_$1_save_LDFLAGS="$LDFLAGS"
|
|
|
|
ompi_check_libnl3_$1_save_LIBS="$LIBS"
|
|
|
|
|
|
|
|
ompi_check_libnl3_$1_orig_CPPFLAGS="$$1_CPPFLAGS"
|
|
|
|
ompi_check_libnl3_$1_orig_LDFLAGS="$$1_LDFLAGS"
|
|
|
|
ompi_check_libnl3_$1_orig_LIBS="$$1_LIBS"
|
|
|
|
|
|
|
|
AS_IF([test "$with_libnl3" != "no"],
|
|
|
|
[AS_IF([test ! -z "$with_libnl3" -a "$with_libnl3" != "yes"],
|
|
|
|
[ompi_check_libnl3_dir="$with_libnl3"])
|
|
|
|
AS_IF([test ! -z "$with_libnl3_libdir" -a "$with_libnl3_libdir" != "yes"],
|
|
|
|
[ompi_check_libnl3_libdir="$with_libnl3_libdir"])
|
|
|
|
|
2014-05-01 18:24:56 +04:00
|
|
|
# OPAL_CHECK_PACKAGE unfortunately can't handle this weird include
|
2014-02-26 11:47:50 +04:00
|
|
|
# dir layout
|
|
|
|
AS_IF([test -n "$ompi_check_libnl3_dir"],
|
|
|
|
[ompi_check_libnl3_includedir="$ompi_check_libnl3_dir/include/libnl3"],
|
|
|
|
[ompi_check_libnl3_includedir="/usr/include/libnl3"])
|
|
|
|
$1_CPPFLAGS="$$1_CPPFLAGS -I$ompi_check_libnl3_includedir"
|
|
|
|
CPPFLAGS="$CPPFLAGS -I$ompi_check_libnl3_includedir"
|
|
|
|
|
|
|
|
AC_CHECK_HEADER([netlink/netlink.h],
|
|
|
|
[# nl_recvmsgs_report appears to be a symbol which
|
|
|
|
# is present in libnl-3 but not libnl (v1)
|
2014-05-01 18:24:56 +04:00
|
|
|
_OPAL_CHECK_PACKAGE_LIB([$1],
|
2014-02-26 11:47:50 +04:00
|
|
|
[nl-3],
|
|
|
|
[nl_recvmsgs_report],
|
|
|
|
[],
|
|
|
|
[$ompi_check_libnl3_dir],
|
|
|
|
[$ompi_check_libnl3_libdir],
|
|
|
|
[ompi_check_libnl3_happy="yes"],
|
|
|
|
[ompi_check_libnl3_happy="no"])],
|
|
|
|
[ompi_check_libnl3_happy=no])
|
|
|
|
|
|
|
|
# make sure that we don't pollute the cache with the results of a
|
|
|
|
# test performed under different CPPFLAGS
|
|
|
|
AS_UNSET([ac_cv_header_netlink_netlink_h])],
|
|
|
|
[ompi_check_libnl3_happy="no"])
|
|
|
|
|
|
|
|
# restore global flags
|
|
|
|
CPPFLAGS="$ompi_check_libnl3_$1_save_CPPFLAGS"
|
|
|
|
LDFLAGS="$ompi_check_libnl3_$1_save_LDFLAGS"
|
|
|
|
LIBS="$ompi_check_libnl3_$1_save_LIBS"
|
|
|
|
|
|
|
|
AS_IF([test "$ompi_check_libnl3_happy" = "yes"],
|
|
|
|
[$2],
|
|
|
|
[AS_IF([test ! -z "$with_libnl3" -a "$with_libnl3" != "no"],
|
|
|
|
[AC_MSG_ERROR([libnl3 support requested but not found. Aborting])])
|
|
|
|
# restore prefixed flags on failure
|
|
|
|
$1_CPPFLAGS="$ompi_check_package_$1_orig_CPPFLAGS"
|
|
|
|
$1_LDFLAGS="$ompi_check_package_$1_orig_LDFLAGS"
|
|
|
|
$1_LIBS="$ompi_check_package_$1_orig_LIBS"
|
|
|
|
$3])
|
|
|
|
])
|
|
|
|
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 04:47:28 +04:00
|
|
|
# MCA_opal_btl_usnic_POST_CONFIG([should_build])
|
2014-02-26 11:47:50 +04:00
|
|
|
# ------------------------------------------
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 04:47:28 +04:00
|
|
|
AC_DEFUN([MCA_opal_btl_usnic_POST_CONFIG], [
|
|
|
|
AM_CONDITIONAL([OPAL_BTL_USNIC_BUILD_UNIT_TESTS],
|
|
|
|
[test "$1" -eq 1 && test "X$enable_opal_btl_usnic_unit_tests" = "Xyes"])
|
|
|
|
AM_CONDITIONAL([OPAL_BTL_USNIC_BUILD_LIBNL1_UTILS],
|
|
|
|
[test "$1" -eq 1 && test "X$enable_opal_btl_usnic_libnl1_utils" = "Xyes"])
|
|
|
|
AM_CONDITIONAL([OPAL_BTL_USNIC_BUILD_LIBNL3_UTILS],
|
|
|
|
[test "$1" -eq 1 && test "X$enable_opal_btl_usnic_libnl3_utils" = "Xyes"])
|
2014-02-26 11:47:50 +04:00
|
|
|
])
|
2013-07-20 02:13:58 +04:00
|
|
|
|
2014-07-30 06:11:04 +04:00
|
|
|
# MCA_btl_usnic_CONFIG([action-if-can-compile],
|
2013-07-20 02:13:58 +04:00
|
|
|
# [action-if-cant-compile])
|
|
|
|
# ------------------------------------------------
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 04:47:28 +04:00
|
|
|
AC_DEFUN([MCA_opal_btl_usnic_CONFIG],[
|
|
|
|
AC_CONFIG_FILES([opal/mca/btl/usnic/Makefile])
|
2013-07-20 02:13:58 +04:00
|
|
|
|
2014-07-30 06:11:04 +04:00
|
|
|
AC_ARG_WITH([usnic],
|
|
|
|
[AS_HELP_STRING([--with-usnic],
|
|
|
|
[If specified, cause an error if usNIC
|
|
|
|
support cannot be built])])
|
|
|
|
|
|
|
|
# If --without-usnic was specified, then gracefully exit.
|
|
|
|
# Otherwise, do the rest of the config.
|
|
|
|
AS_IF([test "x$with_usnic" = "xno"],
|
|
|
|
[AC_MSG_WARN([--without-usnic specified; skipping usnic BTL])
|
|
|
|
$2],
|
|
|
|
[OMPI_BTL_USNIC_DO_CONFIG($1, $2)])
|
|
|
|
])
|
|
|
|
|
|
|
|
AC_DEFUN([OMPI_BTL_USNIC_DO_CONFIG],[
|
2014-02-26 11:47:50 +04:00
|
|
|
# see README.test for information about this scheme
|
|
|
|
AC_ARG_ENABLE([ompi-btl-usnic-unit-tests],
|
|
|
|
[AS_HELP_STRING([--enable-ompi-btl-usnic-unit-tests],
|
|
|
|
[build unit tests for the usnic BTL,
|
|
|
|
including the test runner program,
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 04:47:28 +04:00
|
|
|
opal_btl_usnic_run_tests])])
|
|
|
|
AS_IF([test "X$enable_opal_btl_usnic_unit_tests" = "Xyes"],
|
2014-02-26 11:47:50 +04:00
|
|
|
[unit_tests=1
|
|
|
|
AC_MSG_NOTICE([enabling usnic BTL unit tests])],
|
|
|
|
[unit_tests=0])
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 04:47:28 +04:00
|
|
|
AC_DEFINE_UNQUOTED([OPAL_BTL_USNIC_UNIT_TESTS], [$unit_tests],
|
2014-02-26 11:47:50 +04:00
|
|
|
[define to 1 if usnic BTL unit tests are enabled, 0 otherwise])
|
|
|
|
unset unit_tests
|
|
|
|
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 04:47:28 +04:00
|
|
|
OPAL_CHECK_OPENFABRICS([btl_usnic],
|
2013-07-20 02:13:58 +04:00
|
|
|
[btl_usnic_happy="yes"],
|
|
|
|
[btl_usnic_happy="no"])
|
|
|
|
|
2014-08-04 23:34:55 +04:00
|
|
|
# The current logic in btl_usnic_compat.h checks the OPAL version as a
|
|
|
|
# proxy for the top-level OMPI version. Unfortunately this does the wrong
|
|
|
|
# thing for other top-level projects that might use the usnic BTL, such as
|
|
|
|
# ORCM. ORCM's versioning is totally unrelated to OMPI's. As a short term
|
|
|
|
# workaround, just disqualify ourselves if the OPAL version seems too old.
|
|
|
|
# In the longer term we should be doing something else, like versioning
|
|
|
|
# OPAL and OMPI separately.
|
|
|
|
AS_IF([test "$btl_usnic_happy" = "yes"],
|
|
|
|
[AS_IF([test "$OPAL_MAJOR_VERSION" -eq "1" && \
|
2014-08-06 02:31:07 +04:00
|
|
|
test "$OPAL_MINOR_VERSION" -lt "7"],
|
2014-08-04 23:34:55 +04:00
|
|
|
[AC_MSG_NOTICE([OPAL version appears to be too old, disabling the usnic BTL])
|
|
|
|
btl_usnic_happy=no])])
|
|
|
|
|
2014-01-10 02:17:01 +04:00
|
|
|
# We only want to build on 64 bit Linux.
|
2013-07-20 02:13:58 +04:00
|
|
|
AS_IF([test "$btl_usnic_happy" = "yes"],
|
2014-01-10 02:17:01 +04:00
|
|
|
[AC_CHECK_SIZEOF([void *])
|
|
|
|
AC_MSG_CHECKING([for 64 bit Linux])
|
|
|
|
case $host_os in
|
|
|
|
*linux*)
|
|
|
|
AS_IF([test $ac_cv_sizeof_void_p -eq 8],
|
2013-07-20 02:13:58 +04:00
|
|
|
[btl_usnic_happy=yes],
|
|
|
|
[btl_usnic_happy=no])
|
2014-01-10 02:17:01 +04:00
|
|
|
;;
|
|
|
|
*)
|
|
|
|
btl_usnic_happy=no
|
|
|
|
;;
|
|
|
|
esac
|
|
|
|
AC_MSG_RESULT([$btl_usnic_happy])
|
|
|
|
]
|
|
|
|
)
|
2013-07-20 02:13:58 +04:00
|
|
|
|
|
|
|
AS_IF([test "$btl_usnic_happy" = "yes"],
|
2014-02-26 11:44:35 +04:00
|
|
|
[AC_CHECK_DECLS([IBV_EVENT_GID_CHANGE, ibv_event_type_str], [], [],
|
2014-02-11 23:18:29 +04:00
|
|
|
[#include <infiniband/verbs.h>
|
2013-07-20 02:13:58 +04:00
|
|
|
])
|
|
|
|
]
|
|
|
|
)
|
|
|
|
|
2014-02-26 11:50:10 +04:00
|
|
|
# Search for libnl so we can query routing information. We need to
|
|
|
|
# distinguish between v1 and v3.
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 04:47:28 +04:00
|
|
|
enable_opal_btl_usnic_libnl1_utils=no
|
|
|
|
enable_opal_btl_usnic_libnl3_utils=no
|
2014-02-26 11:50:10 +04:00
|
|
|
AS_IF([test "$btl_usnic_happy" = "yes"],
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 04:47:28 +04:00
|
|
|
[OPAL_CHECK_LIBNL3([btl_usnic_libnl],
|
|
|
|
[enable_opal_btl_usnic_libnl3_utils=yes],
|
|
|
|
[enable_opal_btl_usnic_libnl3_utils=no])
|
2014-02-26 11:50:10 +04:00
|
|
|
|
|
|
|
# fall back to libnl1 if libnl3 could not be found
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 04:47:28 +04:00
|
|
|
AS_IF([test "X$enable_opal_btl_usnic_libnl3_utils" = "Xno"],
|
2014-05-01 18:24:56 +04:00
|
|
|
[OPAL_CHECK_PACKAGE([btl_usnic_libnl],
|
2014-02-26 11:50:10 +04:00
|
|
|
[netlink/netlink.h],
|
|
|
|
[nl],
|
|
|
|
[nl_recvmsgs_default],
|
|
|
|
[],
|
|
|
|
[],
|
|
|
|
[],
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 04:47:28 +04:00
|
|
|
[enable_opal_btl_usnic_libnl1_utils=yes],
|
|
|
|
[enable_opal_btl_usnic_libnl1_utils=no])])
|
2014-02-26 11:50:10 +04:00
|
|
|
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 04:47:28 +04:00
|
|
|
AS_IF([test "X$enable_opal_btl_usnic_libnl3_utils" = "Xno" &&
|
|
|
|
test "X$enable_opal_btl_usnic_libnl1_utils" = "Xno"],
|
2014-02-26 11:50:10 +04:00
|
|
|
[AC_MSG_NOTICE([could not find a libnl or libnl-3, disabling the usnic BTL])
|
|
|
|
btl_usnic_happy="no"])
|
|
|
|
|
|
|
|
btl_usnic_CPPFLAGS="$btl_usnic_CPPFLAGS $btl_usnic_libnl_CPPFLAGS"
|
|
|
|
btl_usnic_CFLAGS="$btl_usnic_CFLAGS $btl_usnic_libnl_CFLAGS"
|
|
|
|
btl_usnic_LDFLAGS="$btl_usnic_LDFLAGS $btl_usnic_libnl_LDFLAGS"
|
|
|
|
btl_usnic_LIBS="$btl_usnic_libnl_LIBS $btl_usnic_LIBS"
|
|
|
|
])
|
|
|
|
|
2013-07-20 02:13:58 +04:00
|
|
|
AS_IF([test "$btl_usnic_happy" = "yes"],
|
|
|
|
[btl_usnic_WRAPPER_EXTRA_LDFLAGS="$btl_usnic_LDFLAGS"
|
|
|
|
btl_usnic_WRAPPER_EXTRA_LIBS="$btl_usnic_LIBS"
|
|
|
|
$1],
|
2014-07-30 06:11:04 +04:00
|
|
|
[AS_IF([test "$with_usnic" = "yes"],
|
|
|
|
[AC_MSG_WARN([--with-usnic specified, but usNIC support cannot be built])
|
|
|
|
AC_MSG_ERROR([Cannot continue])],
|
|
|
|
[$2])
|
|
|
|
])
|
|
|
|
|
2013-07-20 02:13:58 +04:00
|
|
|
|
|
|
|
# Substitute in the things needed to build USNIC
|
|
|
|
AC_SUBST([btl_usnic_CPPFLAGS])
|
2013-11-05 02:51:35 +04:00
|
|
|
AC_SUBST([btl_usnic_CFLAGS])
|
2013-07-20 02:13:58 +04:00
|
|
|
AC_SUBST([btl_usnic_LDFLAGS])
|
|
|
|
AC_SUBST([btl_usnic_LIBS])
|
|
|
|
])dnl
|