1
1
openmpi/config/mca_configure.ac

471 строка
15 KiB
Plaintext
Исходник Обычный вид История

# -*- shell-script -*-
#
# Copyright (c) 2004-2005 The Trustees of Indiana University.
# All rights reserved.
# Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
# All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
############################################################################
# Initialization, version number, and other random setup/init stuff
############################################################################
# Init autoconf
AC_INIT(@PARAM_INIT_FILE@)
AC_PREREQ(2.57)
AC_CONFIG_AUX_DIR(@PARAM_CONFIG_AUX_DIR@)
- massive change for module<-->component name fixes throughout the code base. - many (most) mca type names have "component" or "module" in them, as relevant, just to further distinguish the difference between component data/actions and module data/actions. All developers are encouraged to perpetuate this convention when you create types that are specific to a framework, component, or module - did very little to entire framework (just the basics to make it compile) because it's just about to be almost entirely replaced - ditto for io / romio - did not work on elan or ib components; have to commit and then convert those on a different machine with the right libraries and headers - renamed a bunch of *_module.c files to *_component.c and *module*c to *component*c (a few still remain, e.g., ptl/ib, ptl/elan, etc.) - modified autogen/configure/build process to match new filenames (e.g., output static-components.h instead of static-modules.h) - removed DOS-style cr/lf stuff in ns/ns.h - added newline to end of file src/util/numtostr.h - removed some redundant error checking in the top-level topo functions - added a few {} here and there where people "forgot" to put them in for 1 line blocks ;-) - removed a bunch of MPI_* types from mca header files (replaced with corresponding ompi_* types) - all the ptl components had version numbers in their structs; removed - converted a few more elements in the MCA base to use the OBJ interface -- removed some old manual reference counting kruft This commit was SVN r1830.
2004-08-02 04:24:22 +04:00
# Get the version of @MCA_TYPE@ @MCA_COMPONENT_NAME@ that we are installing.
if test -n "@PARAM_VERSION_FILE@" -a -f "@PARAM_VERSION_FILE@"; then
OMPI_GET_VERSION(@PARAM_VERSION_FILE@, @PARAM_VAR_PREFIX@)
AC_SUBST(@PARAM_VAR_PREFIX@_MAJOR_VERSION)
AC_SUBST(@PARAM_VAR_PREFIX@_MINOR_VERSION)
AC_SUBST(@PARAM_VAR_PREFIX@_RELEASE_VERSION)
AC_SUBST(@PARAM_VAR_PREFIX@_ALPHA_VERSION)
AC_SUBST(@PARAM_VAR_PREFIX@_BETA_VERSION)
AC_SUBST(@PARAM_VAR_PREFIX@_SVN_VERSION)
AC_SUBST(@PARAM_VAR_PREFIX@_VERSION)
version_string="version $@PARAM_VAR_PREFIX@_VERSION"
else
version_string=""
fi
#
# Start it up
#
OMPI_CONFIGURE_SETUP
- massive change for module<-->component name fixes throughout the code base. - many (most) mca type names have "component" or "module" in them, as relevant, just to further distinguish the difference between component data/actions and module data/actions. All developers are encouraged to perpetuate this convention when you create types that are specific to a framework, component, or module - did very little to entire framework (just the basics to make it compile) because it's just about to be almost entirely replaced - ditto for io / romio - did not work on elan or ib components; have to commit and then convert those on a different machine with the right libraries and headers - renamed a bunch of *_module.c files to *_component.c and *module*c to *component*c (a few still remain, e.g., ptl/ib, ptl/elan, etc.) - modified autogen/configure/build process to match new filenames (e.g., output static-components.h instead of static-modules.h) - removed DOS-style cr/lf stuff in ns/ns.h - added newline to end of file src/util/numtostr.h - removed some redundant error checking in the top-level topo functions - added a few {} here and there where people "forgot" to put them in for 1 line blocks ;-) - removed a bunch of MPI_* types from mca header files (replaced with corresponding ompi_* types) - all the ptl components had version numbers in their structs; removed - converted a few more elements in the MCA base to use the OBJ interface -- removed some old manual reference counting kruft This commit was SVN r1830.
2004-08-02 04:24:22 +04:00
ompi_show_title "Configuring MCA @MCA_TYPE@ @MCA_COMPONENT_NAME@ component $version_string"
ompi_show_subtitle "Initialization, setup"
#
# Init automake
# The third argument to AM_INIT_AUTOMAKE surpresses the PACKAGE and
# VERSION macors
#
AM_INIT_AUTOMAKE(@PARAM_AM_NAME@, $@PARAM_VAR_PREFIX@_VERSION, 'no')
#
# GNU C and autotools are inconsistent about whether this is defined
# so let's make it true everywhere for now...
#
AC_GNU_SOURCE
- massive change for module<-->component name fixes throughout the code base. - many (most) mca type names have "component" or "module" in them, as relevant, just to further distinguish the difference between component data/actions and module data/actions. All developers are encouraged to perpetuate this convention when you create types that are specific to a framework, component, or module - did very little to entire framework (just the basics to make it compile) because it's just about to be almost entirely replaced - ditto for io / romio - did not work on elan or ib components; have to commit and then convert those on a different machine with the right libraries and headers - renamed a bunch of *_module.c files to *_component.c and *module*c to *component*c (a few still remain, e.g., ptl/ib, ptl/elan, etc.) - modified autogen/configure/build process to match new filenames (e.g., output static-components.h instead of static-modules.h) - removed DOS-style cr/lf stuff in ns/ns.h - added newline to end of file src/util/numtostr.h - removed some redundant error checking in the top-level topo functions - added a few {} here and there where people "forgot" to put them in for 1 line blocks ;-) - removed a bunch of MPI_* types from mca header files (replaced with corresponding ompi_* types) - all the ptl components had version numbers in their structs; removed - converted a few more elements in the MCA base to use the OBJ interface -- removed some old manual reference counting kruft This commit was SVN r1830.
2004-08-02 04:24:22 +04:00
# Setup the top-level MCA component config.h file
AH_TOP([/* -*- c -*-
*
* Copyright (c) 2004-2005 The Trustees of Indiana University.
* All rights reserved.
* Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
* All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*
* Open MPI configuation header file.
- massive change for module<-->component name fixes throughout the code base. - many (most) mca type names have "component" or "module" in them, as relevant, just to further distinguish the difference between component data/actions and module data/actions. All developers are encouraged to perpetuate this convention when you create types that are specific to a framework, component, or module - did very little to entire framework (just the basics to make it compile) because it's just about to be almost entirely replaced - ditto for io / romio - did not work on elan or ib components; have to commit and then convert those on a different machine with the right libraries and headers - renamed a bunch of *_module.c files to *_component.c and *module*c to *component*c (a few still remain, e.g., ptl/ib, ptl/elan, etc.) - modified autogen/configure/build process to match new filenames (e.g., output static-components.h instead of static-modules.h) - removed DOS-style cr/lf stuff in ns/ns.h - added newline to end of file src/util/numtostr.h - removed some redundant error checking in the top-level topo functions - added a few {} here and there where people "forgot" to put them in for 1 line blocks ;-) - removed a bunch of MPI_* types from mca header files (replaced with corresponding ompi_* types) - all the ptl components had version numbers in their structs; removed - converted a few more elements in the MCA base to use the OBJ interface -- removed some old manual reference counting kruft This commit was SVN r1830.
2004-08-02 04:24:22 +04:00
* MCA @MCA_TYPE@: @MCA_COMPONENT_NAME@ component
*/
#ifndef @PARAM_VAR_PREFIX@_CONFIG_H
#define @PARAM_VAR_PREFIX@_CONFIG_H
])
AH_BOTTOM([#endif /* _@PARAM_VAR_PREFIX@_CONFIG_H */])
#
# Generate the version header template -- only if we have a version
#
if test -n "$version_string"; then
# Add the version header template to AC _ CONFIG _ FILES, but only
# if mca_make_configure.pl found that it was listed in
# configure.params and the file exists, or the default file
# exists. If not, then this will be blank (because
# "$version_string" will never be non-empty, so this block is
# meaningless).
# We do this weird Texas two step for two reasons:
# 1) if Autoconf see AC _ CONFIG _ FILES, it assumes that the
# template must exist (and will error during autoconf if it does
# not), even if the AC _ CONFIG _ FILES is in a conditional block
# (like this one)... which actually makes sense. :-) So we only
# put this AC _ CONFIG _ FILES here if we actually have the
# template file (i.e., mca_make_configure.pl puts it here... or
# not).
# 2) AC _ CONFIG _ COMMANDS, on the other hand, does not require
# any Autoconf-compile-time specialness. It can sit here in a
# conditional shell block and be executed or not at run time, so
# there's no need to mca_make_configure.pl to do anything special
# with it.
@WRITE_VERSION_HEADER_TEMPLATE@
# After config.status has run, compare the version header template
# to the version header. If the version header does not exist,
# create it # from the template. If it does already exist, diff
# it and see if # they're different. If they're different, copy
# the template over the # old version. If they're the same, leave
# the original alone so that # we don't distrub any dependencies.
AC_CONFIG_COMMANDS([@MCA_TYPE@-@MCA_COMPONENT_NAME@],
[if test -f "@MCA_TYPE@-@MCA_COMPONENT_NAME@-version.h"; then
diff "@MCA_TYPE@-@MCA_COMPONENT_NAME@-version.h" "@MCA_TYPE@-@MCA_COMPONENT_NAME@-version.h.template" > /dev/null 2>&1
if test "$?" != "0"; then
cp "@MCA_TYPE@-@MCA_COMPONENT_NAME@-version.h.template" "@MCA_TYPE@-@MCA_COMPONENT_NAME@-version.h"
echo "config.status: regenerating @MCA_TYPE@-@MCA_COMPONENT_NAME@-version.h"
else
echo "config.state: @MCA_TYPE@-@MCA_COMPONENT_NAME@-version.h unchanged"
fi
else
cp "@MCA_TYPE@-@MCA_COMPONENT_NAME@-version.h.template" "@MCA_TYPE@-@MCA_COMPONENT_NAME@-version.h"
echo "config.status: creating @MCA_TYPE@-@MCA_COMPONENT_NAME@-version.h"
fi
rm -f @MCA_TYPE@-@MCA_COMPONENT_NAME@-version.h.template])
fi
#
# Do the normal basics of setup:
# - set CLEANFILES
# - figure out the host type
# - set the default prefix
#
OMPI_BASIC_SETUP
#
# Check to see if the user wants this @MCA_TYPE@ to be the default
#
- massive change for module<-->component name fixes throughout the code base. - many (most) mca type names have "component" or "module" in them, as relevant, just to further distinguish the difference between component data/actions and module data/actions. All developers are encouraged to perpetuate this convention when you create types that are specific to a framework, component, or module - did very little to entire framework (just the basics to make it compile) because it's just about to be almost entirely replaced - ditto for io / romio - did not work on elan or ib components; have to commit and then convert those on a different machine with the right libraries and headers - renamed a bunch of *_module.c files to *_component.c and *module*c to *component*c (a few still remain, e.g., ptl/ib, ptl/elan, etc.) - modified autogen/configure/build process to match new filenames (e.g., output static-components.h instead of static-modules.h) - removed DOS-style cr/lf stuff in ns/ns.h - added newline to end of file src/util/numtostr.h - removed some redundant error checking in the top-level topo functions - added a few {} here and there where people "forgot" to put them in for 1 line blocks ;-) - removed a bunch of MPI_* types from mca header files (replaced with corresponding ompi_* types) - all the ptl components had version numbers in their structs; removed - converted a few more elements in the MCA base to use the OBJ interface -- removed some old manual reference counting kruft This commit was SVN r1830.
2004-08-02 04:24:22 +04:00
AC_MSG_CHECKING([if want the @MCA_COMPONENT_NAME@ @MCA_TYPE@ to be the default])
want_default=0
result=no
AC_ARG_WITH(@PROCESSED_MCA_TYPE@,
AC_HELP_STRING([--with-@PROCESSED_MCA_TYPE@=name],
- massive change for module<-->component name fixes throughout the code base. - many (most) mca type names have "component" or "module" in them, as relevant, just to further distinguish the difference between component data/actions and module data/actions. All developers are encouraged to perpetuate this convention when you create types that are specific to a framework, component, or module - did very little to entire framework (just the basics to make it compile) because it's just about to be almost entirely replaced - ditto for io / romio - did not work on elan or ib components; have to commit and then convert those on a different machine with the right libraries and headers - renamed a bunch of *_module.c files to *_component.c and *module*c to *component*c (a few still remain, e.g., ptl/ib, ptl/elan, etc.) - modified autogen/configure/build process to match new filenames (e.g., output static-components.h instead of static-modules.h) - removed DOS-style cr/lf stuff in ns/ns.h - added newline to end of file src/util/numtostr.h - removed some redundant error checking in the top-level topo functions - added a few {} here and there where people "forgot" to put them in for 1 line blocks ;-) - removed a bunch of MPI_* types from mca header files (replaced with corresponding ompi_* types) - all the ptl components had version numbers in their structs; removed - converted a few more elements in the MCA base to use the OBJ interface -- removed some old manual reference counting kruft This commit was SVN r1830.
2004-08-02 04:24:22 +04:00
[if name is "@MCA_COMPONENT_NAME@", the @MCA_COMPONENT_NAME@ @MCA_TYPE@ will be the default]))
if test "$with_@PROCESSED_MCA_TYPE@" = "@MCA_COMPONENT_NAME@"; then
want_default=1
result=yes
fi
AC_DEFINE_UNQUOTED(@PARAM_VAR_PREFIX@_DEFAULT, $want_default,
- massive change for module<-->component name fixes throughout the code base. - many (most) mca type names have "component" or "module" in them, as relevant, just to further distinguish the difference between component data/actions and module data/actions. All developers are encouraged to perpetuate this convention when you create types that are specific to a framework, component, or module - did very little to entire framework (just the basics to make it compile) because it's just about to be almost entirely replaced - ditto for io / romio - did not work on elan or ib components; have to commit and then convert those on a different machine with the right libraries and headers - renamed a bunch of *_module.c files to *_component.c and *module*c to *component*c (a few still remain, e.g., ptl/ib, ptl/elan, etc.) - modified autogen/configure/build process to match new filenames (e.g., output static-components.h instead of static-modules.h) - removed DOS-style cr/lf stuff in ns/ns.h - added newline to end of file src/util/numtostr.h - removed some redundant error checking in the top-level topo functions - added a few {} here and there where people "forgot" to put them in for 1 line blocks ;-) - removed a bunch of MPI_* types from mca header files (replaced with corresponding ompi_* types) - all the ptl components had version numbers in their structs; removed - converted a few more elements in the MCA base to use the OBJ interface -- removed some old manual reference counting kruft This commit was SVN r1830.
2004-08-02 04:24:22 +04:00
[Whether the @MCA_COMPONENT_NAME@ @MCA_TYPE@ is the default @PROCESSED_MCA_TYPE@ or not])
AC_MSG_RESULT([$result])
#
# Part one of libtool magic
#
AM_ENABLE_SHARED
AM_DISABLE_STATIC
#
- massive change for module<-->component name fixes throughout the code base. - many (most) mca type names have "component" or "module" in them, as relevant, just to further distinguish the difference between component data/actions and module data/actions. All developers are encouraged to perpetuate this convention when you create types that are specific to a framework, component, or module - did very little to entire framework (just the basics to make it compile) because it's just about to be almost entirely replaced - ditto for io / romio - did not work on elan or ib components; have to commit and then convert those on a different machine with the right libraries and headers - renamed a bunch of *_module.c files to *_component.c and *module*c to *component*c (a few still remain, e.g., ptl/ib, ptl/elan, etc.) - modified autogen/configure/build process to match new filenames (e.g., output static-components.h instead of static-modules.h) - removed DOS-style cr/lf stuff in ns/ns.h - added newline to end of file src/util/numtostr.h - removed some redundant error checking in the top-level topo functions - added a few {} here and there where people "forgot" to put them in for 1 line blocks ;-) - removed a bunch of MPI_* types from mca header files (replaced with corresponding ompi_* types) - all the ptl components had version numbers in their structs; removed - converted a few more elements in the MCA base to use the OBJ interface -- removed some old manual reference counting kruft This commit was SVN r1830.
2004-08-02 04:24:22 +04:00
# Find which components should be built as run-time loadable components
# Acceptable combinations:
#
# [default -- no option given]
# --enable-mca-dso
# --enable-mca-dso=[.+,]*COMPONENT_TYPE[.+,]*
# --enable-mca-dso=[.+,]*COMPONENT_TYPE-COMPONENT_NAME[.+,]*
# --disable-mca-dso
#
AC_MSG_CHECKING([if want component to be a DSO])
AC_ARG_ENABLE(mca-dso,
AC_HELP_STRING([--enable-mca-dso=LIST],
[comma-separated list of types and/or type-component pairs that will be built as run-time loadable components (as opposed to statically linked in), if supported on this platform. The default is to build all components as DSOs; the --disable-mca-dso(=LIST) form can be used to disable building all or some types/components as DSOs. If LIST is "@MCA_TYPE@-@MCA_COMPONENT_NAME@" or "@MCA_COMPONENT_NAME@", then @MCA_COMPONENT_NAME@ will be compiled as a DSO (if supported on this platform).]))
# Manual conversion of $kind to its generic name (e.g., crmpi->cr,
# crompi->cr).
case "@MCA_TYPE@" in
crmpi)
generic_type="cr"
;;
crompi)
generic_type="cr"
;;
*)
generic_type="@MCA_TYPE@"
;;
esac
- massive change for module<-->component name fixes throughout the code base. - many (most) mca type names have "component" or "module" in them, as relevant, just to further distinguish the difference between component data/actions and module data/actions. All developers are encouraged to perpetuate this convention when you create types that are specific to a framework, component, or module - did very little to entire framework (just the basics to make it compile) because it's just about to be almost entirely replaced - ditto for io / romio - did not work on elan or ib components; have to commit and then convert those on a different machine with the right libraries and headers - renamed a bunch of *_module.c files to *_component.c and *module*c to *component*c (a few still remain, e.g., ptl/ib, ptl/elan, etc.) - modified autogen/configure/build process to match new filenames (e.g., output static-components.h instead of static-modules.h) - removed DOS-style cr/lf stuff in ns/ns.h - added newline to end of file src/util/numtostr.h - removed some redundant error checking in the top-level topo functions - added a few {} here and there where people "forgot" to put them in for 1 line blocks ;-) - removed a bunch of MPI_* types from mca header files (replaced with corresponding ompi_* types) - all the ptl components had version numbers in their structs; removed - converted a few more elements in the MCA base to use the OBJ interface -- removed some old manual reference counting kruft This commit was SVN r1830.
2004-08-02 04:24:22 +04:00
BUILD_@MCA_TYPE@_@MCA_COMPONENT_NAME@_DSO=0
msg=no
if test "$enable_shared" = "no"; then
msg=no
elif test -z "$enable_mca_dso" -o "$enable_mca_dso" = "yes" -o \
"$enable_mca_dso" = "@MCA_TYPE@" -o "$enable_mca_dso" = "$generic_type"; then
- massive change for module<-->component name fixes throughout the code base. - many (most) mca type names have "component" or "module" in them, as relevant, just to further distinguish the difference between component data/actions and module data/actions. All developers are encouraged to perpetuate this convention when you create types that are specific to a framework, component, or module - did very little to entire framework (just the basics to make it compile) because it's just about to be almost entirely replaced - ditto for io / romio - did not work on elan or ib components; have to commit and then convert those on a different machine with the right libraries and headers - renamed a bunch of *_module.c files to *_component.c and *module*c to *component*c (a few still remain, e.g., ptl/ib, ptl/elan, etc.) - modified autogen/configure/build process to match new filenames (e.g., output static-components.h instead of static-modules.h) - removed DOS-style cr/lf stuff in ns/ns.h - added newline to end of file src/util/numtostr.h - removed some redundant error checking in the top-level topo functions - added a few {} here and there where people "forgot" to put them in for 1 line blocks ;-) - removed a bunch of MPI_* types from mca header files (replaced with corresponding ompi_* types) - all the ptl components had version numbers in their structs; removed - converted a few more elements in the MCA base to use the OBJ interface -- removed some old manual reference counting kruft This commit was SVN r1830.
2004-08-02 04:24:22 +04:00
BUILD_@MCA_TYPE@_@MCA_COMPONENT_NAME@_DSO=1
msg=yes
else
ifs_save="$IFS"
IFS="${IFS}$PATH_SEPARATOR,"
for item in $enable_mca_dso; do
- massive change for module<-->component name fixes throughout the code base. - many (most) mca type names have "component" or "module" in them, as relevant, just to further distinguish the difference between component data/actions and module data/actions. All developers are encouraged to perpetuate this convention when you create types that are specific to a framework, component, or module - did very little to entire framework (just the basics to make it compile) because it's just about to be almost entirely replaced - ditto for io / romio - did not work on elan or ib components; have to commit and then convert those on a different machine with the right libraries and headers - renamed a bunch of *_module.c files to *_component.c and *module*c to *component*c (a few still remain, e.g., ptl/ib, ptl/elan, etc.) - modified autogen/configure/build process to match new filenames (e.g., output static-components.h instead of static-modules.h) - removed DOS-style cr/lf stuff in ns/ns.h - added newline to end of file src/util/numtostr.h - removed some redundant error checking in the top-level topo functions - added a few {} here and there where people "forgot" to put them in for 1 line blocks ;-) - removed a bunch of MPI_* types from mca header files (replaced with corresponding ompi_* types) - all the ptl components had version numbers in their structs; removed - converted a few more elements in the MCA base to use the OBJ interface -- removed some old manual reference counting kruft This commit was SVN r1830.
2004-08-02 04:24:22 +04:00
if test "$item" = "@MCA_TYPE@-@MCA_COMPONENT_NAME@" -o \
"$item" = "@MCA_TYPE@" -o "$item" = "$generic_type"; then
- massive change for module<-->component name fixes throughout the code base. - many (most) mca type names have "component" or "module" in them, as relevant, just to further distinguish the difference between component data/actions and module data/actions. All developers are encouraged to perpetuate this convention when you create types that are specific to a framework, component, or module - did very little to entire framework (just the basics to make it compile) because it's just about to be almost entirely replaced - ditto for io / romio - did not work on elan or ib components; have to commit and then convert those on a different machine with the right libraries and headers - renamed a bunch of *_module.c files to *_component.c and *module*c to *component*c (a few still remain, e.g., ptl/ib, ptl/elan, etc.) - modified autogen/configure/build process to match new filenames (e.g., output static-components.h instead of static-modules.h) - removed DOS-style cr/lf stuff in ns/ns.h - added newline to end of file src/util/numtostr.h - removed some redundant error checking in the top-level topo functions - added a few {} here and there where people "forgot" to put them in for 1 line blocks ;-) - removed a bunch of MPI_* types from mca header files (replaced with corresponding ompi_* types) - all the ptl components had version numbers in their structs; removed - converted a few more elements in the MCA base to use the OBJ interface -- removed some old manual reference counting kruft This commit was SVN r1830.
2004-08-02 04:24:22 +04:00
BUILD_@MCA_TYPE@_@MCA_COMPONENT_NAME@_DSO=1
msg=yes
fi
done
IFS="$ifs_save"
fi
AC_MSG_RESULT([$msg])
unset msg
###########################################################################
# Check for compilers and preprocessors
############################################################################
##################################
# Compiler characteristics
##################################
@C_COMPILER_SETUP@
@CXX_COMPILER_SETUP@
##################################
# Find Open MPI header files
##################################
# Are we building as part of the Open MPI source tree, or outside of it?
AC_ARG_WITH(openmpi,
AC_HELP_STRING([--with-openmpi=DIR],
[specify the location of the Open MPI header files and library. The default is to build within a source tree or be able to find the headers/library without additional CPPFLAGS/LDFLAGS. This option is generally only necessary when components are being configured/built outside of an Open MPI source tree]))
# Establish the top-level OMPI directory. If --with-openmpi was not
# specified, then see if we're building a) within an Open MPI source
# tree, or b) can find the Open MPI headers and library file without
# any additional CPPFLAGS/LDFLAGS.
top_ompi_srcdir=
top_ompi_builddir=
found_ompi_headers=
# Split this into multiple tests because we may have multiple
# possibilities here:
#
# - build in the Open MPI source tree
# - build outside of the Open MPI source tree, and have valid
# --with-openmpi
# - build outside of the Open MPI source tree, and not have a valid
# --with-openmpi
AC_MSG_CHECKING([for Open MPI header files])
# First case: we do not have --with-openmpi, so check and see if we're
# building inside the Open MPI source tree
if test -z "$with_openmpi"; then
# Are we within an Open MPI source tree?
if test -f "$srcdir/../../../../config/mca_configure.ac"; then
# This is needed for VPATH builds, so that it will -I the
# appropriate include directory (don't know why automake
# doesn't do this # automatically).
top_ompi_srcdir='$(top_srcdir)/../../../..'
top_ompi_builddir='$(top_builddir)/../../../..'
INCFLAGS='-I$(top_ompi_srcdir)/src -I$(top_ompi_srcdir)/include -I$(top_ompi_builddir)/src'
found_ompi_headers=1
AC_MSG_RESULT([already in Open MPI source tree])
fi
fi
# If we haven't found the headers yet, then we know that we're *not*
# building within the Open MPI source tree.
if test -z "$found_ompi_headers"; then
top_ompi_srcdir=""
top_ompi_builddir="$top_ompi_srcdir"
fi
# Now see if a) we haven't found the headers yet, and b) we have
# --with-openmpi
if test -z "$found_ompi_headers"; then
# We're outside the Open MPI build tree, and we have been provided
# a top-level directory where the header files live.
if test -n "$with_openmpi"; then
if test -d $with_opemnpi ; then
if test -d "$with_openmpi/include/openmpi" -a \
-f "$with_openmpi/include/openmpi/include/ompi.h"; then
INCFLAGS="-I$with_openmpi/include/openmpi -I$with_openmpi/include"
found_ompi_headers=1
AC_MSG_RESULT([$with_openmpi/include/openmpi])
elif test -d "$with_openmpi/openmpi" -a \
-f "$with_openmpi/openmpi/include/ompi.h"; then
INCFLAGS="-I$with_openmpi/openmpi I$with_openmpi/openmpi"
found_ompi_headers=1
AC_MSG_RESULT([$with_openmpi/openmpi])
elif test -f "$with_openmpi/include/ompi.h"; then
INCFLAGS="-I$with_openmpi -I$with_openmpi/.."
found_ompi_headers=1
AC_MSG_RESULT([$with_openmpi])
else
AC_MSG_RESULT([got bogus --with-openmpi value])
AC_MSG_WARN([*** Directory $with_openmpi exists])
AC_MSG_WARN([*** But cannot seem to find Open MPI headers in it])
AC_MSG_WARN([*** Looking elsewhere...])
fi
else
AC_MSG_RESULT([got bogus --with-openmpi value])
AC_MSG_WARN([*** Directory $with_openmpi does not seem to exist])
AC_MSG_WARN([*** Trying to find Open MPI headers without it...])
fi
fi
# If we've gotten this far and haven't found the Open MPI headers
# yet, then just try compiling a C program with some of the Open
# MPI headers and see if they're found (i.e., if they're in the
# preprocessor's default search path)
if test -z "$found_ompi_headers"; then
AC_COMPILE_IFELSE(AC_LANG_PROGRAM([[#include "include/ompi.h"]],
[[]]),
[found_ompi_headers=1])
if test -n "$found_ompi_headers"; then
AC_MSG_RESULT([default preprocessor path])
fi
fi
fi
# If we got this far and haven't found the Open MPI headers, stick a
# fork in us, 'cause we're done.
if test -z "$found_ompi_headers"; then
AC_MSG_WARN([*** Could not find the Open MPI headers])
AC_MSG_ERROR([*** Cannot continue])
fi
AC_SUBST(top_ompi_srcdir)
AC_SUBST(top_ompi_builddir)
##################################
- massive change for module<-->component name fixes throughout the code base. - many (most) mca type names have "component" or "module" in them, as relevant, just to further distinguish the difference between component data/actions and module data/actions. All developers are encouraged to perpetuate this convention when you create types that are specific to a framework, component, or module - did very little to entire framework (just the basics to make it compile) because it's just about to be almost entirely replaced - ditto for io / romio - did not work on elan or ib components; have to commit and then convert those on a different machine with the right libraries and headers - renamed a bunch of *_module.c files to *_component.c and *module*c to *component*c (a few still remain, e.g., ptl/ib, ptl/elan, etc.) - modified autogen/configure/build process to match new filenames (e.g., output static-components.h instead of static-modules.h) - removed DOS-style cr/lf stuff in ns/ns.h - added newline to end of file src/util/numtostr.h - removed some redundant error checking in the top-level topo functions - added a few {} here and there where people "forgot" to put them in for 1 line blocks ;-) - removed a bunch of MPI_* types from mca header files (replaced with corresponding ompi_* types) - all the ptl components had version numbers in their structs; removed - converted a few more elements in the MCA base to use the OBJ interface -- removed some old manual reference counting kruft This commit was SVN r1830.
2004-08-02 04:24:22 +04:00
# @MCA_TYPE@ @MCA_COMPONENT_NAME@ component specific setup
##################################
#
# Unset all *_EXTRA_* flags
#
unset ABORT
for scope in LIBMPI WRAPPER; do
for flags in CFLAGS CXXFLAGS FFLAGS LDFLAGS LIBS; do
str="unset ${scope}_EXTRA_${flags}"
eval $str
done
done
#
# Invoke configure.stub code stubs, if they exist. Have at least one
# statement (well, m4 macro) in each "if" block so that some older sh
# flavors don't complain.
#
if test "$OMPI_WANT_DIST" = "no"; then
OMPI_LOG_MSG([Running configure stub macro])
@CONFIGURE_STUB_MACRO@
else
OMPI_LOG_MSG([Running configure dist stub macro])
@CONFIGURE_DIST_STUB_MACRO@
fi
#
# If any *_EXTRA_* flags were set, save them in $srcdir/post_configure.sh.
# This will be examined by the top-level configure script.
#
create_post_configure_file() {
if test ! -f post_configure.sh; then
cat > post_configure.sh <<EOF
# This file is automatically generated by configure.
# Manual changes will be lost!
# Generated by: `whoami`
# On machine: `hostname`
# On: `date`
EOF
fi
}
rm -f post_configure.sh
if test -n "$ABORT"; then
create_post_configure_file
echo "ABORT=\"$ABORT\"" >> post_configure.sh
AC_MSG_ERROR([*** ABORT flag is set; aborting])
fi
for scope in LIBMPI WRAPPER; do
for flags in CFLAGS CXXFLAGS FFLAGS LDFLAGS LIBS; do
str="found=\$${scope}_EXTRA_${flags}"
eval $str
if test -n "$found"; then
create_post_configure_file
echo "${scope}_EXTRA_${flags}=\"$found\"" >> post_configure.sh
fi
done
done
#
# We delayed doing it until now just so that long include paths don't
# show up in any of the configure output -- purely aesthetic.
#
CPPFLAGS="$CPPFLAGS $INCFLAGS"
CXXCPPFLAGS="$CXXCPPFLAGS $INCFLAGS"
#
# Delayed the substitution of these until now because they may have
# been modified throughout the course of this script.
#
AC_SUBST(CPPFLAGS)
AC_SUBST(CXXCPPFLAGS)
############################################################################
# libtool magic
############################################################################
ompi_show_subtitle "GNU libtool setup"
AM_PROG_LIBTOOL
- massive change for module<-->component name fixes throughout the code base. - many (most) mca type names have "component" or "module" in them, as relevant, just to further distinguish the difference between component data/actions and module data/actions. All developers are encouraged to perpetuate this convention when you create types that are specific to a framework, component, or module - did very little to entire framework (just the basics to make it compile) because it's just about to be almost entirely replaced - ditto for io / romio - did not work on elan or ib components; have to commit and then convert those on a different machine with the right libraries and headers - renamed a bunch of *_module.c files to *_component.c and *module*c to *component*c (a few still remain, e.g., ptl/ib, ptl/elan, etc.) - modified autogen/configure/build process to match new filenames (e.g., output static-components.h instead of static-modules.h) - removed DOS-style cr/lf stuff in ns/ns.h - added newline to end of file src/util/numtostr.h - removed some redundant error checking in the top-level topo functions - added a few {} here and there where people "forgot" to put them in for 1 line blocks ;-) - removed a bunch of MPI_* types from mca header files (replaced with corresponding ompi_* types) - all the ptl components had version numbers in their structs; removed - converted a few more elements in the MCA base to use the OBJ interface -- removed some old manual reference counting kruft This commit was SVN r1830.
2004-08-02 04:24:22 +04:00
AM_CONDITIONAL(OMPI_BUILD_@MCA_TYPE@_@MCA_COMPONENT_NAME@_DSO,
test "$BUILD_@MCA_TYPE@_@MCA_COMPONENT_NAME@_DSO" = "1")
############################################################################
# Party on
############################################################################
ompi_show_subtitle "Final output"
AM_CONFIG_HEADER([@PARAM_CONFIG_HEADER_FILE@])
AC_CONFIG_FILES([@PARAM_CONFIG_FILES@])
AC_OUTPUT