1
1
openmpi/config/ompi_mca.m4

814 строки
27 KiB
Plaintext
Исходник Обычный вид История

dnl -*- shell-script -*-
dnl
dnl Copyright (c) 2004-2005 The Trustees of Indiana University.
dnl All rights reserved.
dnl Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
dnl All rights reserved.
dnl Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
dnl University of Stuttgart. All rights reserved.
dnl Copyright (c) 2004-2005 The Regents of the University of California.
dnl All rights reserved.
dnl $COPYRIGHT$
dnl
dnl Additional copyrights may follow
dnl
dnl $HEADER$
dnl
AC_DEFUN([OMPI_MCA],[
- 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_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]))
AC_ARG_ENABLE(mca-static,
AC_HELP_STRING([--enable-mca-static=LIST],
[comma-separated list of types and/or
type-component pairs that will be built statically
linked into the library. The default (if DSOs are
supported) is to build all components as DSOs.
Enabling a component as static disables it
building as a DSO.]))
AC_ARG_ENABLE(mca-direct,
AC_HELP_STRING([--enable-mca-direct=LIST],
[comma-separated list of type-component pairs that
will be hard coded as the one component to use for
a given component type, saving the (small)
overhead of the component architecture. LIST must
not be empty and implies given component pairs are
build as static components.]))
#
# First, add all the mca-direct components / types into the mca-static
# lists and create a list of component types that are direct compile,
# in the form DIRECT_[type]=[component]
#
AC_MSG_CHECKING([which components should be direct-linked into the library])
if test "$enable_mca_direct" = "yes" ; then
AC_MSG_RESULT([yes])
AC_MSG_ERROR([*** The enable-mca-direct flag requires an explicit list of
*** type-component pairs. For example, --enable-mca-direct=pml-teg,coll-basic])
elif test ! -z "$enable_mca_direct" -a "$enable_mca_direct" != "" ; then
#
# we need to add this into the static list, unless the static list
# is everything
#
if test "$enable_mca_static" = "no" ; then
AC_MSG_WARN([*** Re-enabling static component support for direct call])
enable_mca_static="$enable_mca_direct"
elif test -z "$enable_mca_static" ; then
enable_mca_static="$enable_mca_direct"
elif test "$enable_mca_static" != "yes" ; then
enable_mca_static="$enable_mca_direct,$enable_mca_static"
fi
ifs_save="$IFS"
IFS="${IFS}$PATH_SEPARATOR,"
msg=
for item in $enable_mca_direct; do
type="`echo $item | cut -f1 -d-`"
comp="`echo $item | cut -f2- -d-`"
if test -z $type -o -z $comp ; then
AC_MSG_ERROR([*** The enable-mca-direct flag requires a
*** list of type-component pairs. Invalid input detected.])
else
str="`echo DIRECT_$type=$comp | sed s/-/_/g`"
eval $str
msg="$item $msg"
fi
done
IFS="$ifs_save"
fi
AC_MSG_RESULT([$msg])
unset msg
#
# Second, set the DSO_all and STATIC_all variables. conflict
# resolution (prefer static) is done in the big loop below
#
AC_MSG_CHECKING([which components should be run-time loadable])
if test "$enable_shared" = "no"; then
DSO_all=0
msg=none
elif test -z "$enable_mca_dso" -o "$enable_mca_dso" = "yes"; then
DSO_all=1
msg=all
elif test "$enable_mca_dso" = "no"; then
DSO_all=0
msg=none
else
DSO_all=0
ifs_save="$IFS"
IFS="${IFS}$PATH_SEPARATOR,"
msg=
for item in $enable_mca_dso; do
str="`echo DSO_$item=1 | sed s/-/_/g`"
eval $str
msg="$item $msg"
done
IFS="$ifs_save"
fi
AC_MSG_RESULT([$msg])
unset msg
if test "$enable_shared" = "no"; then
AC_MSG_WARN([*** Shared libraries have been disabled (--disable-shared])
AC_MSG_WARN([*** Building MCA components as DSOs automatically disabled])
fi
AC_MSG_CHECKING([which components should be static])
if test "$enable_mca_static" = "yes"; then
STATIC_all=1
msg=all
elif test -z "$enable_mca_static" -o "$enable_mca_static" = "no"; then
STATIC_all=0
msg=none
else
STATIC_all=0
ifs_save="$IFS"
IFS="${IFS}$PATH_SEPARATOR,"
msg=
for item in $enable_mca_static; do
str="`echo STATIC_$item=1 | sed s/-/_/g`"
eval $str
msg="$item $msg"
done
IFS="$ifs_save"
fi
AC_MSG_RESULT([$msg])
unset msg
# The list of MCA types (it's fixed)
AC_MSG_CHECKING([for MCA types])
found_types="common allocator coll errmgr gpr io iof mpool ns oob pls pml ptl ras rds rmaps rmgr rml soh topo"
AC_MSG_RESULT([$found_types])
- 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 list of all the non-configure MCA components that were found by
# autogen.sh.
- 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
# config/mca_no_configure_components.m4
MCA_FIND_NO_CONFIGURE_COMPONENTS
- 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
# Now determine the configurable components in each of the types. This
# is a little redundant and could be combined into
- 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_FIND_NO_CONFIGURE_COMPONENTS, but we separate it out for clarity.
# The extern statements and array of pointers to the component global
# structs are written to a file for each type that is #include'd in
# the file for each type.
for type in $found_types; do
- massive change for module<-->component name fixes throughout the code base. - many (most) mca type names have "component" or "module" in them, as relevant, just to further distinguish the difference between component data/actions and module data/actions. All developers are encouraged to perpetuate this convention when you create types that are specific to a framework, component, or module - did very little to entire framework (just the basics to make it compile) because it's just about to be almost entirely replaced - ditto for io / romio - did not work on elan or ib components; have to commit and then convert those on a different machine with the right libraries and headers - renamed a bunch of *_module.c files to *_component.c and *module*c to *component*c (a few still remain, e.g., ptl/ib, ptl/elan, etc.) - modified autogen/configure/build process to match new filenames (e.g., output static-components.h instead of static-modules.h) - removed DOS-style cr/lf stuff in ns/ns.h - added newline to end of file src/util/numtostr.h - removed some redundant error checking in the top-level topo functions - added a few {} here and there where people "forgot" to put them in for 1 line blocks ;-) - removed a bunch of MPI_* types from mca header files (replaced with corresponding ompi_* types) - all the ptl components had version numbers in their structs; removed - converted a few more elements in the MCA base to use the OBJ interface -- removed some old manual reference counting kruft This commit was SVN r1830.
2004-08-02 04:24:22 +04:00
all_components=
static_components=
dso_components=
static_ltlibs=
# Ensure that the directory where the #include file is to live
# exists. Need to do this for VPATH builds, because the directory
# may not exist yet. For the "common" type, it's not really a
# component, so it doesn't have a base.
if test "$type" = "common"; then
outdir=src/mca/common
else
outdir=src/mca/$type/base
fi
total_dir="."
for dir_part in `IFS='/\\'; set X $outdir; shift; echo "$[@]"`; do
total_dir=$total_dir/$dir_part
test -d "$total_dir" ||
mkdir "$total_dir" ||
AC_MSG_ERROR([cannot create $total_dir])
done
# Also ensure that the dynamic-mca base directory exists
total_dir="."
dyndir=src/dynamic-mca/$type
for dir_part in `IFS='/\\'; set X $dyndir; shift; echo "$[@]"`; do
total_dir=$total_dir/$dir_part
test -d "$total_dir" ||
mkdir "$total_dir" ||
AC_MSG_ERROR([cannot create $total_dir])
done
# Remove any previous generated #include files.
outfile_real=$outdir/static-components.h
outfile=$outfile_real.new
rm -f $outfile $outfile.struct $outfile.extern \
$outfile.all $outfile.static $outfile.dyanmic
touch $outfile.struct $outfile.extern \
$outfile.all $outfile.static $outfile.dso
# Manual conversion of $type to its generic name (e.g., crmpi->cr,
# crompi->cr).
# JMS Fix this
case "$type" in
crmpi)
generic_type="cr"
;;
crompi)
generic_type="cr"
;;
*)
generic_type="$type"
;;
esac
# set the direct / no direct flag
str="DIRECT_COMPONENT=\$DIRECT_${type}"
eval $str
if test ! -z "$DIRECT_COMPONENT" ; then
str="MCA_${type}_DIRECT_CALL_COMPONENT=$DIRECT_COMPONENT"
eval $str
str="MCA_${type}_DIRECT_CALL=1"
eval $str
else
str="MCA_${type}_DIRECT_CALL_COMPONENT="
eval $str
str="MCA_${type}_DIRECT_CALL=0"
eval $str
fi
- 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
# Iterate through the list of no-configure components
- 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
foo="found_components=\$MCA_${type}_NO_CONFIGURE_SUBDIRS"
eval $foo
- 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
for component in $found_components; do
m=`basename "$component"`
# build if:
# - the component type is direct and we are that component
# - there is no ompi_ignore file
# - there is an ompi_ignore, but there is an empty ompi_unignore
# - there is an ompi_ignore, but username is in ompi_unignore
if test -d $srcdir/$component ; then
# decide if we want the component to be built or not. This
# is spread out because some of the logic is a little complex
# and test's syntax isn't exactly the greatest. We want to
# build the component by default.
want_component=1
if test -f $srcdir/$component/.ompi_ignore ; then
# If there is an ompi_ignore file, don't build
# the component. Note that this decision can be
# overriden by the unignore logic below.
want_component=0
fi
if test -f $srcdir/$component/.ompi_unignore ; then
# if there is an empty ompi_unignore, that is
# equivalent to having your userid in the unignore file.
# If userid is in the file, unignore the ignore file.
if test ! -s $srcdir/$component/.ompi_unignore ; then
want_component=1
elif test ! -z "`grep $USER $srcdir/$component/.ompi_unignore`" ; then
want_component=1
fi
fi
# if this component type is direct and we are not it, we don't want
# to be built. Otherwise, we do want to be built.
if test ! -z "$DIRECT_COMPONENT" ; then
if test "$DIRECT_COMPONENT" = "$m" ; then
want_component=1
else
want_component=0
fi
fi
if test "$want_component" = "1" ; then
ompi_show_subtitle "MCA component $type:$m (no configure script)"
# Remove any possible sym link in the mca-dynamic tree
rm -f src/dynamic-mca/$type/$m
# Now process the component
MCA_PROCESS_COMPONENT(1, 1, $type, $m)
# Note that the AM_CONDITIONAL for this component is set in
# config/mca_no_configure_components.m4 -- which is generated by
# autogen.sh because we cannot have a variable
# AM_CONDITIONAL name (which we need here). Since
# autogen.sh knows the name that is necessary, it just
# generated the AM_CONDITIONAL directly. Here, we fill in
# the variable that is used in that AM_CONDITIONAL.
if test "$compile_mode" = "dso"; then
value=1
else
value=0
fi
foo="BUILD_${type}_${m}_DSO=$value"
eval $foo
# double check that we can build direct if that was requested
# DIRECT_CALL_HEADER *must* be defined by the component
# (in its post configure) if it
# can be direct built, so we use that as a keyword to tell us
# whether the component was successfully setup or not
if test "$DIRECT_COMPONENT" = "$m" -a \
-z "$MCA_${type}_DIRECT_CALL_HEADER" ; then
AC_MSG_ERROR([${type} component ${m} was requested to be directly linked
into libmpi, but does not support such a configuration. Please choose
another ${type} component for direct compilation or allow all components
of type ${type} to be loaded at runtime.])
fi
fi
fi
done
- 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 all configureable components, run their configure scripts,
# etc.
- 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
for component in $srcdir/src/mca/$type/*; do
FOUND=0
HAPPY=0
m="`basename $component`"
# build if:
# - the component type is direct and we are that component
# - there is no ompi_ignore file
# - there is an ompi_ignore, but there is an empty ompi_unignore
# - there is an ompi_ignore, but username is in ompi_unignore
if test -d $component -a -x $component/configure ; then
want_component=1
if test -f $srcdir/$component/.ompi_ignore ; then
want_component=0
fi
if test -f $srcdir/$component/.ompi_unignore ; then
if test ! -s $srcdir/$component/.ompi_unignore ; then
want_component=1
elif test ! -z "`grep $USER $srcdir/$component/.ompi_unignore`" ; then
want_component=1
fi
fi
# if this component type is direct and we are not it, we don't want
# to be built. Otherwise, we do want to be built.
if test ! -z "$DIRECT_COMPONENT" ; then
if test "$DIRECT_COMPONENT" = "$m" ; then
# BWB - need some check in here to make sure component
# can be built direct!
want_component=1
else
want_component=0
fi
fi
if test "$want_component" = "1" ; then
ompi_show_subtitle "MCA component $type:$m (need to configure)"
# We found one!
FOUND=1
# Remove any possible sym link in the mca-dynamic tree
rm -f src/dyanmic-mca/$type/$m
# Configure the component subdirectory
OMPI_CONFIG_SUBDIR([src/mca/$type/$m],
[$ompi_subdir_args],
[HAPPY=1], [HAPPY=0])
fi
fi
- 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
# Process this component
- 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_PROCESS_COMPONENT($FOUND, $HAPPY, $type, $m)
# double check that we can build direct if that was requested
if test "$DIRECT_COMPONENT" = "$m" -a \
-z "$MCA_${type}_DIRECT_CALL_HEADER" ; then
AC_MSG_ERROR([${type} component ${m} was requested to be directly linked
into libmpi, but does not support such a configuration. Please choose
another ${type} component for direct compilation or allow all components
of type ${type} to be loaded at runtime.])
fi
done
# m4 weirdness: must also do the echo after the sort, or we get a
# string with newlines in it
- 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
all_components="`sort $outfile.all`"
all_components="`echo $all_components`"
static_components="`sort $outfile.static`"
static_components="`echo $static_components`"
dso_components="`sort $outfile.dso`"
dso_components="`echo $dso_components`"
rm -f $outfile $outfile.all $outfile.static $outfile.dso
# Create the final .h file that will be included in the type's
# top-level glue. This lists all the static components. We don't
# need to do this for "common".
if test "$type" != "common"; then
cat > $outfile <<EOF
/*
* \$HEADER\$
*/
`cat $outfile.extern`
- massive change for module<-->component name fixes throughout the code base. - many (most) mca type names have "component" or "module" in them, as relevant, just to further distinguish the difference between component data/actions and module data/actions. All developers are encouraged to perpetuate this convention when you create types that are specific to a framework, component, or module - did very little to entire framework (just the basics to make it compile) because it's just about to be almost entirely replaced - ditto for io / romio - did not work on elan or ib components; have to commit and then convert those on a different machine with the right libraries and headers - renamed a bunch of *_module.c files to *_component.c and *module*c to *component*c (a few still remain, e.g., ptl/ib, ptl/elan, etc.) - modified autogen/configure/build process to match new filenames (e.g., output static-components.h instead of static-modules.h) - removed DOS-style cr/lf stuff in ns/ns.h - added newline to end of file src/util/numtostr.h - removed some redundant error checking in the top-level topo functions - added a few {} here and there where people "forgot" to put them in for 1 line blocks ;-) - removed a bunch of MPI_* types from mca header files (replaced with corresponding ompi_* types) - all the ptl components had version numbers in their structs; removed - converted a few more elements in the MCA base to use the OBJ interface -- removed some old manual reference counting kruft This commit was SVN r1830.
2004-08-02 04:24:22 +04:00
const mca_base_component_t *mca_${type}_base_static_components[[]] = {
`cat $outfile.struct`
NULL
};
EOF
# Only replace the header file if a) it doesn't previously
# exist, or b) the contents are different. Do this to not
# trigger recompilation of certain .c files just because the
# timestamp changed on $outfile_real (similar to the way AC
# handles AC_CONFIG_HEADER files).
diff $outfile $outfile_real > /dev/null 2>&1
if test "$?" != "0"; then
mv $outfile $outfile_real
else
rm -f $outfile
fi
fi
rm -f $outfile.struct $outfile.extern
# Save the results for the Makefile.am's. Note the whacky shell
- 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
# script escaping that is necessary because $components may be
# multiple words, but we also need to substitute on ${type}...
- massive change for module<-->component name fixes throughout the code base. - many (most) mca type names have "component" or "module" in them, as relevant, just to further distinguish the difference between component data/actions and module data/actions. All developers are encouraged to perpetuate this convention when you create types that are specific to a framework, component, or module - did very little to entire framework (just the basics to make it compile) because it's just about to be almost entirely replaced - ditto for io / romio - did not work on elan or ib components; have to commit and then convert those on a different machine with the right libraries and headers - renamed a bunch of *_module.c files to *_component.c and *module*c to *component*c (a few still remain, e.g., ptl/ib, ptl/elan, etc.) - modified autogen/configure/build process to match new filenames (e.g., output static-components.h instead of static-modules.h) - removed DOS-style cr/lf stuff in ns/ns.h - added newline to end of file src/util/numtostr.h - removed some redundant error checking in the top-level topo functions - added a few {} here and there where people "forgot" to put them in for 1 line blocks ;-) - removed a bunch of MPI_* types from mca header files (replaced with corresponding ompi_* types) - all the ptl components had version numbers in their structs; removed - converted a few more elements in the MCA base to use the OBJ interface -- removed some old manual reference counting kruft This commit was SVN r1830.
2004-08-02 04:24:22 +04:00
foo="MCA_${type}_ALL_SUBDIRS"'="$all_components"'
eval "$foo"
- 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
foo="MCA_${type}_STATIC_SUBDIRS"'="$static_components"'
eval "$foo"
- massive change for module<-->component name fixes throughout the code base. - many (most) mca type names have "component" or "module" in them, as relevant, just to further distinguish the difference between component data/actions and module data/actions. All developers are encouraged to perpetuate this convention when you create types that are specific to a framework, component, or module - did very little to entire framework (just the basics to make it compile) because it's just about to be almost entirely replaced - ditto for io / romio - did not work on elan or ib components; have to commit and then convert those on a different machine with the right libraries and headers - renamed a bunch of *_module.c files to *_component.c and *module*c to *component*c (a few still remain, e.g., ptl/ib, ptl/elan, etc.) - modified autogen/configure/build process to match new filenames (e.g., output static-components.h instead of static-modules.h) - removed DOS-style cr/lf stuff in ns/ns.h - added newline to end of file src/util/numtostr.h - removed some redundant error checking in the top-level topo functions - added a few {} here and there where people "forgot" to put them in for 1 line blocks ;-) - removed a bunch of MPI_* types from mca header files (replaced with corresponding ompi_* types) - all the ptl components had version numbers in their structs; removed - converted a few more elements in the MCA base to use the OBJ interface -- removed some old manual reference counting kruft This commit was SVN r1830.
2004-08-02 04:24:22 +04:00
foo="MCA_${type}_DSO_SUBDIRS"'="$dso_components"'
eval "$foo"
foo="MCA_${type}_STATIC_LTLIBS"'="$static_ltlibs"'
eval "$foo"
done
- massive change for module<-->component name fixes throughout the code base. - many (most) mca type names have "component" or "module" in them, as relevant, just to further distinguish the difference between component data/actions and module data/actions. All developers are encouraged to perpetuate this convention when you create types that are specific to a framework, component, or module - did very little to entire framework (just the basics to make it compile) because it's just about to be almost entirely replaced - ditto for io / romio - did not work on elan or ib components; have to commit and then convert those on a different machine with the right libraries and headers - renamed a bunch of *_module.c files to *_component.c and *module*c to *component*c (a few still remain, e.g., ptl/ib, ptl/elan, etc.) - modified autogen/configure/build process to match new filenames (e.g., output static-components.h instead of static-modules.h) - removed DOS-style cr/lf stuff in ns/ns.h - added newline to end of file src/util/numtostr.h - removed some redundant error checking in the top-level topo functions - added a few {} here and there where people "forgot" to put them in for 1 line blocks ;-) - removed a bunch of MPI_* types from mca header files (replaced with corresponding ompi_* types) - all the ptl components had version numbers in their structs; removed - converted a few more elements in the MCA base to use the OBJ interface -- removed some old manual reference counting kruft This commit was SVN r1830.
2004-08-02 04:24:22 +04:00
unset foo type m components structs outfile outdir total_dir file \
all_components static_components dso_components static_ltlibs
# Grumble. It seems that AC_SUBST and AC_DEFINE don't let you
# substitue on a variable name that contains a variable (e.g.,
# OMPI_MCA_$type_SUBDIRS). So we have to do this manually. :-(
# Common types
AC_SUBST(MCA_common_ALL_SUBDIRS)
AC_SUBST(MCA_common_STATIC_SUBDIRS)
AC_SUBST(MCA_common_DSO_SUBDIRS)
AC_SUBST(MCA_common_STATIC_LTLIBS)
# ORTE types
AC_SUBST(MCA_oob_ALL_SUBDIRS)
AC_SUBST(MCA_oob_STATIC_SUBDIRS)
AC_SUBST(MCA_oob_DSO_SUBDIRS)
AC_SUBST(MCA_oob_STATIC_LTLIBS)
AC_SUBST(MCA_dps_ALL_SUBDIRS)
AC_SUBST(MCA_dps_STATIC_SUBDIRS)
AC_SUBST(MCA_dps_DSO_SUBDIRS)
AC_SUBST(MCA_dps_STATIC_LTLIBS)
AC_SUBST(MCA_errmgr_ALL_SUBDIRS)
AC_SUBST(MCA_errmgr_STATIC_SUBDIRS)
AC_SUBST(MCA_errmgr_DSO_SUBDIRS)
AC_SUBST(MCA_errmgr_STATIC_LTLIBS)
AC_SUBST(MCA_gpr_ALL_SUBDIRS)
AC_SUBST(MCA_gpr_STATIC_SUBDIRS)
AC_SUBST(MCA_gpr_DSO_SUBDIRS)
AC_SUBST(MCA_gpr_STATIC_LTLIBS)
AC_SUBST(MCA_ns_ALL_SUBDIRS)
AC_SUBST(MCA_ns_STATIC_SUBDIRS)
AC_SUBST(MCA_ns_DSO_SUBDIRS)
AC_SUBST(MCA_ns_STATIC_LTLIBS)
AC_SUBST(MCA_ras_ALL_SUBDIRS)
AC_SUBST(MCA_ras_STATIC_SUBDIRS)
AC_SUBST(MCA_ras_DSO_SUBDIRS)
AC_SUBST(MCA_ras_STATIC_LTLIBS)
AC_SUBST(MCA_rds_ALL_SUBDIRS)
AC_SUBST(MCA_rds_STATIC_SUBDIRS)
AC_SUBST(MCA_rds_DSO_SUBDIRS)
AC_SUBST(MCA_rds_STATIC_LTLIBS)
AC_SUBST(MCA_rmaps_ALL_SUBDIRS)
AC_SUBST(MCA_rmaps_STATIC_SUBDIRS)
AC_SUBST(MCA_rmaps_DSO_SUBDIRS)
AC_SUBST(MCA_rmaps_STATIC_LTLIBS)
AC_SUBST(MCA_rmgr_ALL_SUBDIRS)
AC_SUBST(MCA_rmgr_STATIC_SUBDIRS)
AC_SUBST(MCA_rmgr_DSO_SUBDIRS)
AC_SUBST(MCA_rmgr_STATIC_LTLIBS)
AC_SUBST(MCA_pls_ALL_SUBDIRS)
AC_SUBST(MCA_pls_STATIC_SUBDIRS)
AC_SUBST(MCA_pls_DSO_SUBDIRS)
AC_SUBST(MCA_pls_STATIC_LTLIBS)
AC_SUBST(MCA_rml_ALL_SUBDIRS)
AC_SUBST(MCA_rml_STATIC_SUBDIRS)
AC_SUBST(MCA_rml_DSO_SUBDIRS)
AC_SUBST(MCA_rml_STATIC_LTLIBS)
AC_SUBST(MCA_soh_ALL_SUBDIRS)
AC_SUBST(MCA_soh_STATIC_SUBDIRS)
AC_SUBST(MCA_soh_DSO_SUBDIRS)
AC_SUBST(MCA_soh_STATIC_LTLIBS)
AC_SUBST(MCA_svc_ALL_SUBDIRS)
AC_SUBST(MCA_svc_STATIC_SUBDIRS)
AC_SUBST(MCA_svc_DSO_SUBDIRS)
AC_SUBST(MCA_svc_STATIC_LTLIBS)
# MPI types
AC_SUBST(MCA_allocator_ALL_SUBDIRS)
AC_SUBST(MCA_allocator_STATIC_SUBDIRS)
AC_SUBST(MCA_allocator_DSO_SUBDIRS)
AC_SUBST(MCA_allocator_STATIC_LTLIBS)
AC_SUBST(MCA_coll_ALL_SUBDIRS)
AC_SUBST(MCA_coll_STATIC_SUBDIRS)
AC_SUBST(MCA_coll_DSO_SUBDIRS)
AC_SUBST(MCA_coll_STATIC_LTLIBS)
AC_SUBST(MCA_io_ALL_SUBDIRS)
AC_SUBST(MCA_io_STATIC_SUBDIRS)
AC_SUBST(MCA_io_DSO_SUBDIRS)
AC_SUBST(MCA_io_STATIC_LTLIBS)
AC_SUBST(MCA_iof_ALL_SUBDIRS)
AC_SUBST(MCA_iof_STATIC_SUBDIRS)
AC_SUBST(MCA_iof_DSO_SUBDIRS)
AC_SUBST(MCA_iof_STATIC_LTLIBS)
AC_SUBST(MCA_mpool_ALL_SUBDIRS)
AC_SUBST(MCA_mpool_STATIC_SUBDIRS)
AC_SUBST(MCA_mpool_DSO_SUBDIRS)
AC_SUBST(MCA_mpool_STATIC_LTLIBS)
AC_SUBST(MCA_pml_ALL_SUBDIRS)
AC_SUBST(MCA_pml_STATIC_SUBDIRS)
AC_SUBST(MCA_pml_DSO_SUBDIRS)
AC_SUBST(MCA_pml_STATIC_LTLIBS)
OMPI_SETUP_DIRECT_CALL(pml)
AC_SUBST(MCA_ptl_ALL_SUBDIRS)
AC_SUBST(MCA_ptl_STATIC_SUBDIRS)
AC_SUBST(MCA_ptl_DSO_SUBDIRS)
AC_SUBST(MCA_ptl_STATIC_LTLIBS)
AC_SUBST(MCA_schema_ALL_SUBDIRS)
AC_SUBST(MCA_schema_STATIC_SUBDIRS)
AC_SUBST(MCA_schema_DSO_SUBDIRS)
AC_SUBST(MCA_schema_STATIC_LTLIBS)
AC_SUBST(MCA_topo_ALL_SUBDIRS)
AC_SUBST(MCA_topo_STATIC_SUBDIRS)
AC_SUBST(MCA_topo_DSO_SUBDIRS)
AC_SUBST(MCA_topo_STATIC_LTLIBS)
# Finally, now that we've filled in all the test variables, get all
- 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
# the AM_CONDITIONALs that indicate whether to build components as shared
# or 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
# config/mca_no_configure_components.m4
MCA_AMC_NO_CONFIGURE_COMPONENTS])
dnl -----------------------------------------------------------------------
dnl -----------------------------------------------------------------------
dnl -----------------------------------------------------------------------
- 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_DEFUN([MCA_PROCESS_COMPONENT],[
FOUND=$1
HAPPY=$2
type=$3
m=$4
# See if it dropped an output file for us to pick up some
# shell variables in.
infile="src/mca/$type/$m/post_configure.sh"
- massive change for module<-->component name fixes throughout the code base. - many (most) mca type names have "component" or "module" in them, as relevant, just to further distinguish the difference between component data/actions and module data/actions. All developers are encouraged to perpetuate this convention when you create types that are specific to a framework, component, or module - did very little to entire framework (just the basics to make it compile) because it's just about to be almost entirely replaced - ditto for io / romio - did not work on elan or ib components; have to commit and then convert those on a different machine with the right libraries and headers - renamed a bunch of *_module.c files to *_component.c and *module*c to *component*c (a few still remain, e.g., ptl/ib, ptl/elan, etc.) - modified autogen/configure/build process to match new filenames (e.g., output static-components.h instead of static-modules.h) - removed DOS-style cr/lf stuff in ns/ns.h - added newline to end of file src/util/numtostr.h - removed some redundant error checking in the top-level topo functions - added a few {} here and there where people "forgot" to put them in for 1 line blocks ;-) - removed a bunch of MPI_* types from mca header files (replaced with corresponding ompi_* types) - all the ptl components had version numbers in their structs; removed - converted a few more elements in the MCA base to use the OBJ interface -- removed some old manual reference counting kruft This commit was SVN r1830.
2004-08-02 04:24:22 +04:00
# Did we find a valid component, and did its configure run
# successfully?
if test "$HAPPY" = "1"; 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
# Add this subdir to the mast list of all MCA component subdirs
echo $m >> $outfile.all
- massive change for module<-->component name fixes throughout the code base. - many (most) mca type names have "component" or "module" in them, as relevant, just to further distinguish the difference between component data/actions and module data/actions. All developers are encouraged to perpetuate this convention when you create types that are specific to a framework, component, or module - did very little to entire framework (just the basics to make it compile) because it's just about to be almost entirely replaced - ditto for io / romio - did not work on elan or ib components; have to commit and then convert those on a different machine with the right libraries and headers - renamed a bunch of *_module.c files to *_component.c and *module*c to *component*c (a few still remain, e.g., ptl/ib, ptl/elan, etc.) - modified autogen/configure/build process to match new filenames (e.g., output static-components.h instead of static-modules.h) - removed DOS-style cr/lf stuff in ns/ns.h - added newline to end of file src/util/numtostr.h - removed some redundant error checking in the top-level topo functions - added a few {} here and there where people "forgot" to put them in for 1 line blocks ;-) - removed a bunch of MPI_* types from mca header files (replaced with corresponding ompi_* types) - all the ptl components had version numbers in their structs; removed - converted a few more elements in the MCA base to use the OBJ interface -- removed some old manual reference counting kruft This commit was SVN r1830.
2004-08-02 04:24:22 +04:00
# Is this component going to built staic or shared?
str="SHARED_TYPE=\$DSO_$type"
eval $str
str="SHARED_GENERIC_TYPE=\$DSO_$generic_type"
eval $str
- massive change for module<-->component name fixes throughout the code base. - many (most) mca type names have "component" or "module" in them, as relevant, just to further distinguish the difference between component data/actions and module data/actions. All developers are encouraged to perpetuate this convention when you create types that are specific to a framework, component, or module - did very little to entire framework (just the basics to make it compile) because it's just about to be almost entirely replaced - ditto for io / romio - did not work on elan or ib components; have to commit and then convert those on a different machine with the right libraries and headers - renamed a bunch of *_module.c files to *_component.c and *module*c to *component*c (a few still remain, e.g., ptl/ib, ptl/elan, etc.) - modified autogen/configure/build process to match new filenames (e.g., output static-components.h instead of static-modules.h) - removed DOS-style cr/lf stuff in ns/ns.h - added newline to end of file src/util/numtostr.h - removed some redundant error checking in the top-level topo functions - added a few {} here and there where people "forgot" to put them in for 1 line blocks ;-) - removed a bunch of MPI_* types from mca header files (replaced with corresponding ompi_* types) - all the ptl components had version numbers in their structs; removed - converted a few more elements in the MCA base to use the OBJ interface -- removed some old manual reference counting kruft This commit was SVN r1830.
2004-08-02 04:24:22 +04:00
str="SHARED_COMPONENT=\$DSO_${type}_$m"
eval $str
str="STATIC_TYPE=\$STATIC_$type"
eval $str
str="STATIC_GENERIC_TYPE=\$STATIC_$generic_type"
eval $str
str="STATIC_COMPONENT=\$STATIC_${type}_$m"
eval $str
shared_mode_override=static
# Setup for either shared or static
if test "$STATIC_TYPE" = "1" -o \
"$STATIC_GENERIC_TYPE" = "1" -o \
"$STATIC_COMPONENT" = "1" -o \
"$STATIC_all" = "1" ; then
compile_mode="static"
elif test "$shared_mode_override" = "dso" -o \
"$SHARED_TYPE" = "1" -o \
"$SHARED_GENERIC_TYPE" = "1" -o \
"$SHARED_COMPONENT" = "1" -o \
"$DSO_all" = "1"; then
compile_mode="dso"
else
compile_mode="static"
fi
if test "$compile_mode" = "dso" ; then
echo $m >> $outfile.dso
rm -f "src/dynamic-mca/$type/$m"
$LN_S "$OMPI_TOP_BUILDDIR/src/mca/$type/$m" \
"src/dynamic-mca/$type/$m"
else
static_ltlibs="mca/$type/$m/libmca_${type}_${m}.la $static_ltlibs"
echo "extern const mca_base_component_t mca_${type}_${m}_component;" >> $outfile.extern
echo " &mca_${type}_${m}_component, " >> $outfile.struct
echo $m >> $outfile.static
fi
# Output pretty results
- 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 MCA component $type:$m can compile])
AC_MSG_RESULT([yes])
- 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([for MCA component $type:$m compile mode])
if test "$DIRECT_COMPONENT" = "$m" ; then
AC_MSG_RESULT([$compile_mode - direct])
else
AC_MSG_RESULT([$compile_mode])
fi
# If there's an output file, add the values to
# scope_EXTRA_flags.
if test -f $infile; then
# First check for the ABORT tag
line="`grep ABORT= $infile | cut -d= -f2-`"
if test -n "$line" -a "$line" != "no"; then
AC_MSG_WARN([MCA component configure script told me to abort])
AC_MSG_ERROR([cannot continue])
fi
# If we're not compiling statically, then only take the
# "ALWAYS" tags (a uniq will be performed at the end -- no
# need to worry about duplicate flags here)
for flags in LDFLAGS LIBS; do
var_in="LIBMPI_ALWAYS_EXTRA_${flags}"
var_out="LIBMPI_EXTRA_${flags}"
line="`grep $var_in= $infile | cut -d= -f2-`"
eval "line=$line"
if test -n "$line"; then
str="$var_out="'"$'"$var_out $var_in $line"'"'
eval $str
fi
done
for flags in CFLAGS CXXFLAGS FFLAGS FCFLAGS LDFLAGS LIBS; do
var_in="WRAPPER_ALWAYS_EXTRA_${flags}"
var_out="WRAPPER_EXTRA_${flags}"
line="`grep $var_in= $infile | cut -d= -f2-`"
eval "line=$line"
if test -n "$line"; then
str="$var_out="'"$'"$var_out $var_in $line"'"'
eval $str
fi
done
# Check for flags passed up from the component. If we're
# compiling statically, then take all flags passed up from the
# component.
if test "$compile_mode" = "static"; then
for flags in LDFLAGS LIBS; do
var="LIBMPI_EXTRA_${flags}"
line="`grep $var= $infile | cut -d= -f2-`"
eval "line=$line"
if test -n "$line"; then
str="$var="'"$'"$var $line"'"'
eval $str
fi
done
for flags in CFLAGS CXXFLAGS FFLAGS FCFLAGS LDFLAGS LIBS; do
var="WRAPPER_EXTRA_${flags}"
line="`grep $var= $infile | cut -d= -f2-`"
eval "line=$line"
if test -n "$line"; then
str="$var="'"$'"$var $line"'"'
eval $str
fi
done
fi
dnl check for direct call header to include. This will be
dnl AC_SUBSTed later.
if test "$DIRECT_COMPONENT" = "$m" ; then
if test "`grep DIRECT_CALL_HEADER $infile`" != "" ; then
line="`grep DIRECT_CALL_HEADER $infile | cut -d= -f2-`"
str="MCA_${type}_DIRECT_CALL_HEADER=\\\"$line\\\""
eval $str
else
AC_MSG_ERROR([*** ${type} component ${m} was supposed to be direct-called, but
*** does not appear to support direct calling.
*** Aborting])
fi
fi
else
# were we supposed to have found something in the
# post_configure.sh, but the file didn't exist?
if test "$DIRECT_COMPONENT" = "$m" ; then
AC_MSG_ERROR([*** ${type} component ${m} was supposed to be direct-called, but
*** does not appear to support direct calling.
*** Aborting])
fi
fi
elif test "$FOUND" = "1"; 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
AC_MSG_CHECKING([if MCA component $type:$m can compile])
AC_MSG_RESULT([no])
- 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 this component was requested as the default for this
# type, then abort.
str="foo="'"$'"with_$type"'"'
eval $str
str="bar="'"$'"with_$generic_type"'"'
eval $str
if test "$foo" = "$m" -o "$bar" = "$m"; then
AC_MSG_WARN([MCA component "$m" failed to configure properly])
AC_MSG_WARN([This component was selected as the default])
AC_MSG_ERROR([Cannot continue])
exit 1
fi
fi
])
AC_DEFUN([OMPI_SETUP_DIRECT_CALL],[
AC_SUBST(MCA_$1_DIRECT_CALL_HEADER)
AC_DEFINE_UNQUOTED([MCA_$1_DIRECT_CALL], [$MCA_$1_DIRECT_CALL],
[Defined to 1 if $1 should use direct calls instead of components])
AC_DEFINE_UNQUOTED([MCA_$1_DIRECT_CALL_COMPONENT], [$MCA_$1_DIRECT_CALL_COMPONENT],
[name of component to use for direct calls, if MCA_$1_DIRECT_CALL is 1])
OMPI_WRITE_DIRECT_CALL_HEADER($1)
])
AC_DEFUN([OMPI_WRITE_DIRECT_CALL_HEADER],[
AC_CONFIG_FILES(src/mca/$1/$1_direct_call.h.template)
AC_CONFIG_COMMANDS($1-direct,
[if test -f "src/mca/$1/$1_direct_call"; then
diff "src/mca/$1/$1_direct_call.h" "src/mca/$1/$1_direct_call.h.template" > /dev/null 2>&1
if test "$?" != "0"; then
cp "src/mca/$1/$1_direct_call.h.template" "src/mca/$1/$1_direct_call.h"
echo "config.status: regenerating src/mca/$1/$1_direct_call.h"
else
echo "config.state: src/mca/$1/$1_direct_call.h unchanged"
fi
else
cp "src/mca/$1/$1_direct_call.h.template" "src/mca/$1/$1_direct_call.h"
echo "config.status: creating src/mca/$1/$1_direct_call.h"
fi
rm src/mca/$1/$1_direct_call.h.template])
])