1
1
openmpi/configure.ac

1465 строки
46 KiB
Plaintext
Исходник Обычный вид История

2003-11-22 19:36:58 +03:00
# -*- shell-script -*-
#
# Copyright (c) 2004-2009 The Trustees of Indiana University and Indiana
# University Research and Technology
# Corporation. All rights reserved.
# Copyright (c) 2004-2005 The University of Tennessee and The University
# of Tennessee Research Foundation. All rights
# reserved.
# Copyright (c) 2004-2007 High Performance Computing Center Stuttgart,
# University of Stuttgart. All rights reserved.
# Copyright (c) 2004-2005 The Regents of the University of California.
# All rights reserved.
- As proposed in RFC and telcon, warn the user about deprecated functionality (per MPI-2.1). This warning can be toggled using --enable-mpi-interface-warning (default OFF), but can be selectively turned on passing mpicc -DOMPI_WANT_MPI_INTERFACE_WARNING Using icc, gcc < 4.5, warnings (such as in mpi2basic_tests) show: type_vector.c:83: warning: ‘MPI_Type_hvector’ is deprecated (declared at /home/../usr/include/mpi.h:1379) Using gcc-4.5 (gcc-svn) these show up as: type_vector.c:83: warning: ‘MPI_Type_hvector’ is deprecated (declared at /home/../usr/include/mpi.h:1379): MPI_Type_hvector is superseded by MPI_Type_create_hvector in MPI-2.0 Jeff and I propose to turn such warnings on with Open MPI-1.7 by default. - Detection of user-level compiler is handled using the preprocessor checks of GASnet's other/portable_platform.h (thanks to Paul Hargrove and Dan Bonachea) adapted into ompi/include/mpi_portable_platform.h (see comments). The OMPI-build time detection is output (Familyname and Version) with ompi_info. This functionality (actually any upcoming __attribute__) are turned off, if a different compiler (and version) is being detected. - Note, that any warnings regarding (user-compiler!=build-compiler) as discussed in the RFC are _not_ included for now. - Tested on Linux with --enable-mpi-interface-warning on Linux, gcc-4.5 (deprecated w/ specific msg) Linux, gcc-4.3 (deprecated w/o specific msg) Linux, pathscale 3.1 (deprecated w/o specific msg) Linux, icc-11.0 (deprecated w/o specific msg) Linux, PGI-8.0.6 accepts __deprecated__ but does not issue a warning, further investigation needed... This commit was SVN r21262.
2009-05-22 08:39:43 +04:00
# Copyright (c) 2006-2009 Cisco Systems, Inc. All rights reserved.
# Copyright (c) 2006-2008 Sun Microsystems, Inc. All rights reserved.
# Copyright (c) 2006-2007 Los Alamos National Security, LLC. All rights
# reserved.
- As proposed in RFC and telcon, warn the user about deprecated functionality (per MPI-2.1). This warning can be toggled using --enable-mpi-interface-warning (default OFF), but can be selectively turned on passing mpicc -DOMPI_WANT_MPI_INTERFACE_WARNING Using icc, gcc < 4.5, warnings (such as in mpi2basic_tests) show: type_vector.c:83: warning: ‘MPI_Type_hvector’ is deprecated (declared at /home/../usr/include/mpi.h:1379) Using gcc-4.5 (gcc-svn) these show up as: type_vector.c:83: warning: ‘MPI_Type_hvector’ is deprecated (declared at /home/../usr/include/mpi.h:1379): MPI_Type_hvector is superseded by MPI_Type_create_hvector in MPI-2.0 Jeff and I propose to turn such warnings on with Open MPI-1.7 by default. - Detection of user-level compiler is handled using the preprocessor checks of GASnet's other/portable_platform.h (thanks to Paul Hargrove and Dan Bonachea) adapted into ompi/include/mpi_portable_platform.h (see comments). The OMPI-build time detection is output (Familyname and Version) with ompi_info. This functionality (actually any upcoming __attribute__) are turned off, if a different compiler (and version) is being detected. - Note, that any warnings regarding (user-compiler!=build-compiler) as discussed in the RFC are _not_ included for now. - Tested on Linux with --enable-mpi-interface-warning on Linux, gcc-4.5 (deprecated w/ specific msg) Linux, gcc-4.3 (deprecated w/o specific msg) Linux, pathscale 3.1 (deprecated w/o specific msg) Linux, icc-11.0 (deprecated w/o specific msg) Linux, PGI-8.0.6 accepts __deprecated__ but does not issue a warning, further investigation needed... This commit was SVN r21262.
2009-05-22 08:39:43 +04:00
# Copyright (c) 2009 Oak Ridge National Labs. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
2003-11-22 19:36:58 +03:00
#
############################################################################
# Initialization, version number, and other random setup/init stuff
############################################################################
# Init autoconf
# We don't have the version number to put in here yet, and we can't
# call OMPI_GET_VERSION (etc.) before AC_INIT. So use the shell
# version.
AC_INIT([Open MPI],
[m4_normalize(esyscmd([config/ompi_get_version.sh VERSION --base]))],
[http://www.open-mpi.org/community/help/], [openmpi])
AC_PREREQ(2.60)
AC_CONFIG_AUX_DIR(config)
AC_CONFIG_MACRO_DIR(config)
2003-11-22 19:36:58 +03:00
# Get our platform support file. This has to be done very, very early
# because it twiddles random bits of autoconf
OMPI_LOAD_PLATFORM
#
# Init automake
#
AM_INIT_AUTOMAKE([foreign dist-bzip2 subdir-objects no-define 1.10 tar-ustar])
m4_pattern_allow([m4_ifval])
# Make configure depend on the VERSION file, since it's used in AC_INIT
AC_SUBST([CONFIGURE_DEPENDENCIES], ['$(top_srcdir)/VERSION'])
#
# Start it up
#
OMPI_CONFIGURE_SETUP
ompi_show_title "Configuring Open MPI"
ompi_show_subtitle "Checking versions"
# Get the version of OMPI that we are installing
OMPI_SAVE_VERSION([OMPI], [Open MPI], [$srcdir/VERSION],
[ompi/include/ompi/version.h])
2006-03-12 07:43:59 +03:00
OMPI_SAVE_VERSION([ORTE], [Open Run-Time Environment], [$srcdir/VERSION],
[orte/include/orte/version.h])
OMPI_SAVE_VERSION([OPAL], [Open Portable Access Layer], [$srcdir/VERSION],
[opal/include/opal/version.h])
# Get shared library version numbers
. $srcdir/VERSION
AC_SUBST(libmpi_so_version)
AC_SUBST(libmpi_cxx_so_version)
AC_SUBST(libmpi_f77_so_version)
AC_SUBST(libmpi_f90_so_version)
AC_SUBST(libopen_rte_so_version)
AC_SUBST(libopen_pal_so_version)
# List header files to generate
- As proposed in RFC and telcon, warn the user about deprecated functionality (per MPI-2.1). This warning can be toggled using --enable-mpi-interface-warning (default OFF), but can be selectively turned on passing mpicc -DOMPI_WANT_MPI_INTERFACE_WARNING Using icc, gcc < 4.5, warnings (such as in mpi2basic_tests) show: type_vector.c:83: warning: ‘MPI_Type_hvector’ is deprecated (declared at /home/../usr/include/mpi.h:1379) Using gcc-4.5 (gcc-svn) these show up as: type_vector.c:83: warning: ‘MPI_Type_hvector’ is deprecated (declared at /home/../usr/include/mpi.h:1379): MPI_Type_hvector is superseded by MPI_Type_create_hvector in MPI-2.0 Jeff and I propose to turn such warnings on with Open MPI-1.7 by default. - Detection of user-level compiler is handled using the preprocessor checks of GASnet's other/portable_platform.h (thanks to Paul Hargrove and Dan Bonachea) adapted into ompi/include/mpi_portable_platform.h (see comments). The OMPI-build time detection is output (Familyname and Version) with ompi_info. This functionality (actually any upcoming __attribute__) are turned off, if a different compiler (and version) is being detected. - Note, that any warnings regarding (user-compiler!=build-compiler) as discussed in the RFC are _not_ included for now. - Tested on Linux with --enable-mpi-interface-warning on Linux, gcc-4.5 (deprecated w/ specific msg) Linux, gcc-4.3 (deprecated w/o specific msg) Linux, pathscale 3.1 (deprecated w/o specific msg) Linux, icc-11.0 (deprecated w/o specific msg) Linux, PGI-8.0.6 accepts __deprecated__ but does not issue a warning, further investigation needed... This commit was SVN r21262.
2009-05-22 08:39:43 +04:00
AM_CONFIG_HEADER([opal/include/opal_config.h orte/include/orte_config.h ompi/include/ompi_config.h ompi/include/mpi.h ompi/include/mpi_portable_platform.h])
# override/fixup the version numbers set by AC_INIT, since on
# developer builds, there's no good way to know what the version is
# before running configure :(. We only use the base version number
# (ie, no svn r numbers) for the version set in AC_INIT. This will
# always match reality because we add the VERSION file (the only way
# to change the major.minor.release{greek}) into the configure
# dependencies. PACKAGE_VERSION the AC_DEFINE doesn't change once set
# the first time -- AC_INIT's input (so it doesn't have an r number in
# it). PACKAGE_VERSION the AC_SUBST can be rewritten along the way,
# and we'd like it to have the r number in it so that it shows up in
# the tarball name, so it is set to the full version here.
PACKAGE_VERSION="$OPAL_VERSION"
PACKAGE_STRING="${PACKAGE_NAME} ${PACKAGE_VERSION}"
VERSION="${PACKAGE_VERSION}"
ompi_show_subtitle "Initialization, setup"
2003-11-22 19:36:58 +03:00
OMPI_TOP_BUILDDIR="`pwd`"
AC_SUBST(OMPI_TOP_BUILDDIR)
2003-11-22 19:36:58 +03:00
cd "$srcdir"
OMPI_TOP_SRCDIR="`pwd`"
AC_SUBST(OMPI_TOP_SRCDIR)
cd "$OMPI_TOP_BUILDDIR"
2003-11-22 19:36:58 +03:00
AC_MSG_NOTICE([builddir: $OMPI_TOP_BUILDDIR])
AC_MSG_NOTICE([srcdir: $OMPI_TOP_SRCDIR])
if test "$OMPI_TOP_BUILDDIR" != "$OMPI_TOP_SRCDIR"; then
2003-11-22 19:36:58 +03:00
AC_MSG_NOTICE([Detected VPATH build])
fi
# Setup the top of the opal/include/opal_config.h file
2003-11-22 19:36:58 +03:00
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 (c) 2004-2005 High Performance Computing Center Stuttgart,
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
2003-11-22 19:36:58 +03:00
*
* Function: - OS, CPU and compiler dependent configuration
2003-11-22 19:36:58 +03:00
*/
#ifndef OPAL_CONFIG_H
#define OPAL_CONFIG_H
])
AH_BOTTOM([
#include "opal_config_bottom.h"
#endif /* OPAL_CONFIG_H */
2003-11-22 19:36:58 +03:00
])
- 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
# Other basic setup stuff (shared with components)
2003-11-22 19:36:58 +03:00
OMPI_BASIC_SETUP
2003-11-22 19:36:58 +03:00
top_ompi_srcdir="$OMPI_TOP_SRCDIR"
AC_SUBST(top_ompi_srcdir)
top_ompi_builddir="$OMPI_TOP_BUILDDIR"
AC_SUBST(top_ompi_builddir)
2003-11-22 19:36:58 +03:00
############################################################################
# Configuration options
############################################################################
OPAL_CONFIGURE_OPTIONS
ORTE_CONFIGURE_OPTIONS
OMPI_CONFIGURE_OPTIONS
2003-11-22 19:36:58 +03:00
if test "$enable_binaries" = "no" -a "$enable_dist" = "yes"; then
AC_MSG_WARN([--disable-binaries is incompatible with --enable dist])
AC_MSG_ERROR([Cannot continue])
fi
2003-11-22 19:36:58 +03:00
############################################################################
# Libtool: part one
# (before C compiler setup)
############################################################################
#
# Part one of libtool magic. Enable static so that we have the --with
# tests done up here and can check for OS. Save the values of
# $enable_static and $enable_shared before setting the defaults,
# because if the user specified --[en|dis]able-[static|shared] on the
# command line, they'll already be set. In this way, we can tell if
# the user requested something or if the default was set here.
#
ompi_enable_shared="$enable_shared"
ompi_enable_static="$enable_static"
AM_ENABLE_SHARED
AM_DISABLE_STATIC
2003-11-22 19:36:58 +03:00
OMPI_SETUP_WRAPPER_INIT
##################################
# Check for known incompatibility
##################################
# Do *not* print a message that we're checking the OS because this
# test is *not* meant to be an all-inclusive "if it passes this test,
# then configure must succeed" test. This test is *only* mean to
# screen out the versions of OS X where we know OMPI will cause kernel
# panics because of bad implementations of pty's. See
# https://svn.open-mpi.org/trac/ompi/ticket/1637 for details.
# We do not support OS X before version 10.4 (Tiger)
case $host_os in
# Corresponds to OS X 10.0 - 10.3 (additional [] quoting for m4)
darwin[[4567]]*)
AC_MSG_WARN([Open MPI does not support OS X prior to version 10.4 (Tiger)])
AC_MSG_ERROR([Cannot continue])
esac
2003-11-22 19:36:58 +03:00
############################################################################
# Check for compilers and preprocessors
############################################################################
ompi_show_title "Compiler and preprocessor tests"
2003-11-22 19:36:58 +03:00
##################################
# C compiler characteristics
##################################
OMPI_SETUP_CC
2003-11-22 19:36:58 +03:00
# If we build on a windows environment with the windows compiler and linker
# then we need some translation functions from the opal/win32 directory.
AM_CONDITIONAL(OMPI_NEED_WINDOWS_REPLACEMENTS,
test "$ompi_cv_c_compiler_vendor" = "microsoft" )
# Do all Interix detections if necessary
OMPI_INTERIX
# Does the compiler support "ident"-like constructs?
OMPI_CHECK_IDENT([CC], [CFLAGS], [c], [C])
#
# Check for some types
#
AC_CHECK_TYPES(int8_t)
AC_CHECK_TYPES(uint8_t)
AC_CHECK_TYPES(int16_t)
AC_CHECK_TYPES(uint16_t)
AC_CHECK_TYPES(int32_t)
AC_CHECK_TYPES(uint32_t)
AC_CHECK_TYPES(int64_t)
AC_CHECK_TYPES(uint64_t)
- Split the datatype engine into two parts: an MPI specific part in OMPI and a language agnostic part in OPAL. The convertor is completely moved into OPAL. This offers several benefits as described in RFC http://www.open-mpi.org/community/lists/devel/2009/07/6387.php namely: - Fewer basic types (int* and float* types, boolean and wchar - Fixing naming scheme to ompi-nomenclature. - Usability outside of the ompi-layer. - Due to the fixed nature of simple opal types, their information is completely known at compile time and therefore constified - With fewer datatypes (22), the actual sizes of bit-field types may be reduced from 64 to 32 bits, allowing reorganizing the opal_datatype structure, eliminating holes and keeping data required in convertor (upon send/recv) in one cacheline... This has implications to the convertor-datastructure and other parts of the code. - Several performance tests have been run, the netpipe latency does not change with this patch on Linux/x86-64 on the smoky cluster. - Extensive tests have been done to verify correctness (no new regressions) using: 1. mpi_test_suite on linux/x86-64 using clean ompi-trunk and ompi-ddt: a. running both trunk and ompi-ddt resulted in no differences (except for MPI_SHORT_INT and MPI_TYPE_MIX_LB_UB do now run correctly). b. with --enable-memchecker and running under valgrind (one buglet when run with static found in test-suite, commited) 2. ibm testsuite on linux/x86-64 using clean ompi-trunk and ompi-ddt: all passed (except for the dynamic/ tests failed!! as trunk/MTT) 3. compilation and usage of HDF5 tests on Jaguar using PGI and PathScale compilers. 4. compilation and usage on Scicortex. - Please note, that for the heterogeneous case, (-m32 compiled binaries/ompi), neither ompi-trunk, nor ompi-ddt branch would successfully launch. This commit was SVN r21641.
2009-07-13 08:56:31 +04:00
AC_CHECK_TYPES(int128_t)
AC_CHECK_TYPES(uint128_t)
AC_CHECK_TYPES(long long)
AC_CHECK_TYPES(long double)
AC_CHECK_TYPES(float _Complex)
AC_CHECK_TYPES(double _Complex)
AC_CHECK_TYPES(long double _Complex)
AC_CHECK_TYPES(intptr_t)
AC_CHECK_TYPES(uintptr_t)
AC_CHECK_TYPES(mode_t)
AC_CHECK_TYPES(ssize_t)
AC_CHECK_TYPES(ptrdiff_t)
#
# Check for type sizes
#
AC_CHECK_SIZEOF(char)
AC_CHECK_SIZEOF(short)
AC_CHECK_SIZEOF(int)
AC_CHECK_SIZEOF(long)
if test $ac_cv_type_long_long = yes; then
AC_CHECK_SIZEOF(long long)
fi
- Split the datatype engine into two parts: an MPI specific part in OMPI and a language agnostic part in OPAL. The convertor is completely moved into OPAL. This offers several benefits as described in RFC http://www.open-mpi.org/community/lists/devel/2009/07/6387.php namely: - Fewer basic types (int* and float* types, boolean and wchar - Fixing naming scheme to ompi-nomenclature. - Usability outside of the ompi-layer. - Due to the fixed nature of simple opal types, their information is completely known at compile time and therefore constified - With fewer datatypes (22), the actual sizes of bit-field types may be reduced from 64 to 32 bits, allowing reorganizing the opal_datatype structure, eliminating holes and keeping data required in convertor (upon send/recv) in one cacheline... This has implications to the convertor-datastructure and other parts of the code. - Several performance tests have been run, the netpipe latency does not change with this patch on Linux/x86-64 on the smoky cluster. - Extensive tests have been done to verify correctness (no new regressions) using: 1. mpi_test_suite on linux/x86-64 using clean ompi-trunk and ompi-ddt: a. running both trunk and ompi-ddt resulted in no differences (except for MPI_SHORT_INT and MPI_TYPE_MIX_LB_UB do now run correctly). b. with --enable-memchecker and running under valgrind (one buglet when run with static found in test-suite, commited) 2. ibm testsuite on linux/x86-64 using clean ompi-trunk and ompi-ddt: all passed (except for the dynamic/ tests failed!! as trunk/MTT) 3. compilation and usage of HDF5 tests on Jaguar using PGI and PathScale compilers. 4. compilation and usage on Scicortex. - Please note, that for the heterogeneous case, (-m32 compiled binaries/ompi), neither ompi-trunk, nor ompi-ddt branch would successfully launch. This commit was SVN r21641.
2009-07-13 08:56:31 +04:00
AC_CHECK_SIZEOF(float)
AC_CHECK_SIZEOF(double)
if test $ac_cv_type_long_double = yes; then
AC_CHECK_SIZEOF(long double)
fi
- Split the datatype engine into two parts: an MPI specific part in OMPI and a language agnostic part in OPAL. The convertor is completely moved into OPAL. This offers several benefits as described in RFC http://www.open-mpi.org/community/lists/devel/2009/07/6387.php namely: - Fewer basic types (int* and float* types, boolean and wchar - Fixing naming scheme to ompi-nomenclature. - Usability outside of the ompi-layer. - Due to the fixed nature of simple opal types, their information is completely known at compile time and therefore constified - With fewer datatypes (22), the actual sizes of bit-field types may be reduced from 64 to 32 bits, allowing reorganizing the opal_datatype structure, eliminating holes and keeping data required in convertor (upon send/recv) in one cacheline... This has implications to the convertor-datastructure and other parts of the code. - Several performance tests have been run, the netpipe latency does not change with this patch on Linux/x86-64 on the smoky cluster. - Extensive tests have been done to verify correctness (no new regressions) using: 1. mpi_test_suite on linux/x86-64 using clean ompi-trunk and ompi-ddt: a. running both trunk and ompi-ddt resulted in no differences (except for MPI_SHORT_INT and MPI_TYPE_MIX_LB_UB do now run correctly). b. with --enable-memchecker and running under valgrind (one buglet when run with static found in test-suite, commited) 2. ibm testsuite on linux/x86-64 using clean ompi-trunk and ompi-ddt: all passed (except for the dynamic/ tests failed!! as trunk/MTT) 3. compilation and usage of HDF5 tests on Jaguar using PGI and PathScale compilers. 4. compilation and usage on Scicortex. - Please note, that for the heterogeneous case, (-m32 compiled binaries/ompi), neither ompi-trunk, nor ompi-ddt branch would successfully launch. This commit was SVN r21641.
2009-07-13 08:56:31 +04:00
if test "$ac_cv_type_float__Complex" = yes; then
AC_CHECK_SIZEOF(float _Complex)
fi
if test "$ac_cv_type_double__Complex" = yes; then
AC_CHECK_SIZEOF(double _Complex)
fi
if test "$ac_cv_type_long_double__Complex" = yes; then
AC_CHECK_SIZEOF(long double _Complex)
fi
AC_CHECK_SIZEOF(void *)
AC_CHECK_SIZEOF(size_t)
if test $ac_cv_type_ssize_t = yes ; then
AC_CHECK_SIZEOF(ssize_t)
fi
if test $ac_cv_type_ptrdiff_t = yes; then
AC_CHECK_SIZEOF(ptrdiff_t)
fi
- Split the datatype engine into two parts: an MPI specific part in OMPI and a language agnostic part in OPAL. The convertor is completely moved into OPAL. This offers several benefits as described in RFC http://www.open-mpi.org/community/lists/devel/2009/07/6387.php namely: - Fewer basic types (int* and float* types, boolean and wchar - Fixing naming scheme to ompi-nomenclature. - Usability outside of the ompi-layer. - Due to the fixed nature of simple opal types, their information is completely known at compile time and therefore constified - With fewer datatypes (22), the actual sizes of bit-field types may be reduced from 64 to 32 bits, allowing reorganizing the opal_datatype structure, eliminating holes and keeping data required in convertor (upon send/recv) in one cacheline... This has implications to the convertor-datastructure and other parts of the code. - Several performance tests have been run, the netpipe latency does not change with this patch on Linux/x86-64 on the smoky cluster. - Extensive tests have been done to verify correctness (no new regressions) using: 1. mpi_test_suite on linux/x86-64 using clean ompi-trunk and ompi-ddt: a. running both trunk and ompi-ddt resulted in no differences (except for MPI_SHORT_INT and MPI_TYPE_MIX_LB_UB do now run correctly). b. with --enable-memchecker and running under valgrind (one buglet when run with static found in test-suite, commited) 2. ibm testsuite on linux/x86-64 using clean ompi-trunk and ompi-ddt: all passed (except for the dynamic/ tests failed!! as trunk/MTT) 3. compilation and usage of HDF5 tests on Jaguar using PGI and PathScale compilers. 4. compilation and usage on Scicortex. - Please note, that for the heterogeneous case, (-m32 compiled binaries/ompi), neither ompi-trunk, nor ompi-ddt branch would successfully launch. This commit was SVN r21641.
2009-07-13 08:56:31 +04:00
AC_CHECK_SIZEOF(wchar_t)
#
# Check for type alignments
#
- Split the datatype engine into two parts: an MPI specific part in OMPI and a language agnostic part in OPAL. The convertor is completely moved into OPAL. This offers several benefits as described in RFC http://www.open-mpi.org/community/lists/devel/2009/07/6387.php namely: - Fewer basic types (int* and float* types, boolean and wchar - Fixing naming scheme to ompi-nomenclature. - Usability outside of the ompi-layer. - Due to the fixed nature of simple opal types, their information is completely known at compile time and therefore constified - With fewer datatypes (22), the actual sizes of bit-field types may be reduced from 64 to 32 bits, allowing reorganizing the opal_datatype structure, eliminating holes and keeping data required in convertor (upon send/recv) in one cacheline... This has implications to the convertor-datastructure and other parts of the code. - Several performance tests have been run, the netpipe latency does not change with this patch on Linux/x86-64 on the smoky cluster. - Extensive tests have been done to verify correctness (no new regressions) using: 1. mpi_test_suite on linux/x86-64 using clean ompi-trunk and ompi-ddt: a. running both trunk and ompi-ddt resulted in no differences (except for MPI_SHORT_INT and MPI_TYPE_MIX_LB_UB do now run correctly). b. with --enable-memchecker and running under valgrind (one buglet when run with static found in test-suite, commited) 2. ibm testsuite on linux/x86-64 using clean ompi-trunk and ompi-ddt: all passed (except for the dynamic/ tests failed!! as trunk/MTT) 3. compilation and usage of HDF5 tests on Jaguar using PGI and PathScale compilers. 4. compilation and usage on Scicortex. - Please note, that for the heterogeneous case, (-m32 compiled binaries/ompi), neither ompi-trunk, nor ompi-ddt branch would successfully launch. This commit was SVN r21641.
2009-07-13 08:56:31 +04:00
OMPI_C_GET_ALIGNMENT(_Bool, OPAL_ALIGNMENT_BOOL)
OMPI_C_GET_ALIGNMENT(int8_t, OPAL_ALIGNMENT_INT8)
OMPI_C_GET_ALIGNMENT(int16_t, OPAL_ALIGNMENT_INT16)
OMPI_C_GET_ALIGNMENT(int32_t, OPAL_ALIGNMENT_INT32)
OMPI_C_GET_ALIGNMENT(int64_t, OPAL_ALIGNMENT_INT64)
if test "$ac_cv_type_int128_t" = yes ; then
OMPI_C_GET_ALIGNMENT(int128_t, OPAL_ALIGNMENT_INT128)
fi
OMPI_C_GET_ALIGNMENT(char, OPAL_ALIGNMENT_CHAR)
OMPI_C_GET_ALIGNMENT(short, OPAL_ALIGNMENT_SHORT)
- Split the datatype engine into two parts: an MPI specific part in OMPI and a language agnostic part in OPAL. The convertor is completely moved into OPAL. This offers several benefits as described in RFC http://www.open-mpi.org/community/lists/devel/2009/07/6387.php namely: - Fewer basic types (int* and float* types, boolean and wchar - Fixing naming scheme to ompi-nomenclature. - Usability outside of the ompi-layer. - Due to the fixed nature of simple opal types, their information is completely known at compile time and therefore constified - With fewer datatypes (22), the actual sizes of bit-field types may be reduced from 64 to 32 bits, allowing reorganizing the opal_datatype structure, eliminating holes and keeping data required in convertor (upon send/recv) in one cacheline... This has implications to the convertor-datastructure and other parts of the code. - Several performance tests have been run, the netpipe latency does not change with this patch on Linux/x86-64 on the smoky cluster. - Extensive tests have been done to verify correctness (no new regressions) using: 1. mpi_test_suite on linux/x86-64 using clean ompi-trunk and ompi-ddt: a. running both trunk and ompi-ddt resulted in no differences (except for MPI_SHORT_INT and MPI_TYPE_MIX_LB_UB do now run correctly). b. with --enable-memchecker and running under valgrind (one buglet when run with static found in test-suite, commited) 2. ibm testsuite on linux/x86-64 using clean ompi-trunk and ompi-ddt: all passed (except for the dynamic/ tests failed!! as trunk/MTT) 3. compilation and usage of HDF5 tests on Jaguar using PGI and PathScale compilers. 4. compilation and usage on Scicortex. - Please note, that for the heterogeneous case, (-m32 compiled binaries/ompi), neither ompi-trunk, nor ompi-ddt branch would successfully launch. This commit was SVN r21641.
2009-07-13 08:56:31 +04:00
OMPI_C_GET_ALIGNMENT(wchar_t, OPAL_ALIGNMENT_WCHAR)
OMPI_C_GET_ALIGNMENT(int, OPAL_ALIGNMENT_INT)
OMPI_C_GET_ALIGNMENT(long, OPAL_ALIGNMENT_LONG)
- Split the datatype engine into two parts: an MPI specific part in OMPI and a language agnostic part in OPAL. The convertor is completely moved into OPAL. This offers several benefits as described in RFC http://www.open-mpi.org/community/lists/devel/2009/07/6387.php namely: - Fewer basic types (int* and float* types, boolean and wchar - Fixing naming scheme to ompi-nomenclature. - Usability outside of the ompi-layer. - Due to the fixed nature of simple opal types, their information is completely known at compile time and therefore constified - With fewer datatypes (22), the actual sizes of bit-field types may be reduced from 64 to 32 bits, allowing reorganizing the opal_datatype structure, eliminating holes and keeping data required in convertor (upon send/recv) in one cacheline... This has implications to the convertor-datastructure and other parts of the code. - Several performance tests have been run, the netpipe latency does not change with this patch on Linux/x86-64 on the smoky cluster. - Extensive tests have been done to verify correctness (no new regressions) using: 1. mpi_test_suite on linux/x86-64 using clean ompi-trunk and ompi-ddt: a. running both trunk and ompi-ddt resulted in no differences (except for MPI_SHORT_INT and MPI_TYPE_MIX_LB_UB do now run correctly). b. with --enable-memchecker and running under valgrind (one buglet when run with static found in test-suite, commited) 2. ibm testsuite on linux/x86-64 using clean ompi-trunk and ompi-ddt: all passed (except for the dynamic/ tests failed!! as trunk/MTT) 3. compilation and usage of HDF5 tests on Jaguar using PGI and PathScale compilers. 4. compilation and usage on Scicortex. - Please note, that for the heterogeneous case, (-m32 compiled binaries/ompi), neither ompi-trunk, nor ompi-ddt branch would successfully launch. This commit was SVN r21641.
2009-07-13 08:56:31 +04:00
if test "$ac_cv_type_long_long" = yes; then
OMPI_C_GET_ALIGNMENT(long long, OPAL_ALIGNMENT_LONG_LONG)
fi
- Split the datatype engine into two parts: an MPI specific part in OMPI and a language agnostic part in OPAL. The convertor is completely moved into OPAL. This offers several benefits as described in RFC http://www.open-mpi.org/community/lists/devel/2009/07/6387.php namely: - Fewer basic types (int* and float* types, boolean and wchar - Fixing naming scheme to ompi-nomenclature. - Usability outside of the ompi-layer. - Due to the fixed nature of simple opal types, their information is completely known at compile time and therefore constified - With fewer datatypes (22), the actual sizes of bit-field types may be reduced from 64 to 32 bits, allowing reorganizing the opal_datatype structure, eliminating holes and keeping data required in convertor (upon send/recv) in one cacheline... This has implications to the convertor-datastructure and other parts of the code. - Several performance tests have been run, the netpipe latency does not change with this patch on Linux/x86-64 on the smoky cluster. - Extensive tests have been done to verify correctness (no new regressions) using: 1. mpi_test_suite on linux/x86-64 using clean ompi-trunk and ompi-ddt: a. running both trunk and ompi-ddt resulted in no differences (except for MPI_SHORT_INT and MPI_TYPE_MIX_LB_UB do now run correctly). b. with --enable-memchecker and running under valgrind (one buglet when run with static found in test-suite, commited) 2. ibm testsuite on linux/x86-64 using clean ompi-trunk and ompi-ddt: all passed (except for the dynamic/ tests failed!! as trunk/MTT) 3. compilation and usage of HDF5 tests on Jaguar using PGI and PathScale compilers. 4. compilation and usage on Scicortex. - Please note, that for the heterogeneous case, (-m32 compiled binaries/ompi), neither ompi-trunk, nor ompi-ddt branch would successfully launch. This commit was SVN r21641.
2009-07-13 08:56:31 +04:00
OMPI_C_GET_ALIGNMENT(float, OPAL_ALIGNMENT_FLOAT)
OMPI_C_GET_ALIGNMENT(double, OPAL_ALIGNMENT_DOUBLE)
if test "$ac_cv_type_long_double" = yes; then
OMPI_C_GET_ALIGNMENT(long double, OPAL_ALIGNMENT_LONG_DOUBLE)
fi
- Split the datatype engine into two parts: an MPI specific part in OMPI and a language agnostic part in OPAL. The convertor is completely moved into OPAL. This offers several benefits as described in RFC http://www.open-mpi.org/community/lists/devel/2009/07/6387.php namely: - Fewer basic types (int* and float* types, boolean and wchar - Fixing naming scheme to ompi-nomenclature. - Usability outside of the ompi-layer. - Due to the fixed nature of simple opal types, their information is completely known at compile time and therefore constified - With fewer datatypes (22), the actual sizes of bit-field types may be reduced from 64 to 32 bits, allowing reorganizing the opal_datatype structure, eliminating holes and keeping data required in convertor (upon send/recv) in one cacheline... This has implications to the convertor-datastructure and other parts of the code. - Several performance tests have been run, the netpipe latency does not change with this patch on Linux/x86-64 on the smoky cluster. - Extensive tests have been done to verify correctness (no new regressions) using: 1. mpi_test_suite on linux/x86-64 using clean ompi-trunk and ompi-ddt: a. running both trunk and ompi-ddt resulted in no differences (except for MPI_SHORT_INT and MPI_TYPE_MIX_LB_UB do now run correctly). b. with --enable-memchecker and running under valgrind (one buglet when run with static found in test-suite, commited) 2. ibm testsuite on linux/x86-64 using clean ompi-trunk and ompi-ddt: all passed (except for the dynamic/ tests failed!! as trunk/MTT) 3. compilation and usage of HDF5 tests on Jaguar using PGI and PathScale compilers. 4. compilation and usage on Scicortex. - Please note, that for the heterogeneous case, (-m32 compiled binaries/ompi), neither ompi-trunk, nor ompi-ddt branch would successfully launch. This commit was SVN r21641.
2009-07-13 08:56:31 +04:00
if test "$ac_cv_type_float__Complex" = yes; then
OMPI_C_GET_ALIGNMENT(float _Complex, OPAL_ALIGNMENT_FLOAT_COMPLEX)
fi
if test "$ac_cv_type_double__Complex" = yes; then
OMPI_C_GET_ALIGNMENT(double _Complex, OPAL_ALIGNMENT_DOUBLE_COMPLEX)
fi
if test "$ac_cv_type_long_double__Complex" = yes; then
OMPI_C_GET_ALIGNMENT(long double _Complex, OPAL_ALIGNMENT_LONG_DOUBLE_COMPLEX)
fi
OMPI_C_GET_ALIGNMENT(void *, OPAL_ALIGNMENT_VOID_P)
#
# Does the C compiler native support "bool"? (i.e., without
# <stdbool.h> or any other help)
#
AC_MSG_CHECKING(for C bool type)
AC_COMPILE_IFELSE(AC_LANG_PROGRAM([[
AC_INCLUDES_DEFAULT]],
[[bool bar, foo = true; bar = foo;]]),
[OPAL_NEED_C_BOOL=0 MSG=yes],[OPAL_NEED_C_BOOL=1 MSG=no])
AC_DEFINE_UNQUOTED(OPAL_NEED_C_BOOL, $OPAL_NEED_C_BOOL,
[Whether the C compiler supports "bool" without any other help (such as <stdbool.h>)])
AC_MSG_RESULT([$MSG])
AC_CHECK_SIZEOF(_Bool)
#
# Check for other compiler characteristics
#
if test "$GCC" = "yes"; then
# gcc 2.96 will emit oodles of warnings if you use "inline" with
# -pedantic (which we do in developer builds). However,
# "__inline__" is ok. So we have to force gcc to select the
# right one. If you use -pedantic, the AC_C_INLINE test will fail
# (because it names a function foo() -- without the (void)). So
# we turn off all the picky flags, turn on -ansi mode (which is
# implied by -pedantic), and set warnings to be errors. Hence,
# this does the following (for 2.96):
#
# - causes the check for "inline" to emit a warning, which then
# fails
# - checks for __inline__, which then emits no error, and works
#
# This also works nicely for gcc 3.x because "inline" will work on
# the first check, and all is fine. :-)
CFLAGS_save="$CFLAGS"
CFLAGS="$OMPI_CFLAGS_BEFORE_PICKY -Werror -ansi"
fi
AC_C_INLINE
# Microsoft compilers support 2 versions of restrict. One for functions, and
# one for variables. The problem is that they don't have an equivalent
# syntax, and the autoconf restrict detection is unable to detect them
# correctly. It detect the restrict keyword as __restrict which break the
# rules for function syntax which is declspec(restrict).
if test "x$ompi_cv_c_compiler_vendor" != "xmicrosoft"; then
AC_C_RESTRICT
fi
OMPI_C_WEAK_SYMBOLS
if test "$GCC" = "yes"; then
CFLAGS="$CFLAGS_save"
fi
2003-11-22 19:36:58 +03:00
if test "x$CC" = "xicc"; then
OMPI_CHECK_ICC_VARARGS
fi
# If we want the profiling layer:
# - If the C compiler has weak symbols, use those.
# - If not, then set to compile the code again with #define's in a
# separate directory.
if test "$WANT_WEAK_SYMBOLS" = "0"; then
OPAL_C_HAVE_WEAK_SYMBOLS=0
fi
if test "$WANT_MPI_PROFILING" = "1"; then
if test "$OPAL_C_HAVE_WEAK_SYMBOLS" = "1"; then
OMPI_PROFILING_COMPILE_SEPARATELY=0
else
OMPI_PROFILING_COMPILE_SEPARATELY=1
fi
else
OMPI_PROFILING_COMPILE_SEPARATELY=0
fi
# Check if we support the offsetof compiler directive
OPAL_CHECK_OFFSETOF
#
# There are 2 layers to the MPI Language binidings One layer generates
# MPI_* bindings. The other layer generates PMPI_* bindings. The
# following conditions determine whether each (or both) these layers
# are built.
# 1. MPI_* bindings are needed if:
# - Profiling is not required
# - Profiling is required but weak symbols are not
# supported
# 2. PMPI_* bindings are needed if profiling is required. Hence we
# define 2 conditionals which tell us whether each of these layers
# need to be built or NOT
#
AM_CONDITIONAL(WANT_MPI_BINDINGS_LAYER,
test "$WANT_MPI_PROFILING" = 0 -o "$OMPI_PROFILING_COMPILE_SEPARATELY" = 1)
AM_CONDITIONAL(WANT_PMPI_BINDINGS_LAYER,
test "$WANT_MPI_PROFILING" = 1)
AM_CONDITIONAL(COMPILE_PROFILING_SEPARATELY,
test "$OMPI_PROFILING_COMPILE_SEPARATELY" = 1)
AC_DEFINE_UNQUOTED(OMPI_ENABLE_MPI_PROFILING, $WANT_MPI_PROFILING,
[Whether we want MPI profiling or not])
AC_DEFINE_UNQUOTED(OPAL_HAVE_WEAK_SYMBOLS, $OPAL_C_HAVE_WEAK_SYMBOLS,
[Whether we have weak symbols or not])
2003-11-22 19:36:58 +03:00
##################################
# C++ compiler characteristics
##################################
OMPI_SETUP_CXX
2003-11-22 19:36:58 +03:00
# check if we want C++ support
AM_CONDITIONAL(WANT_MPI_CXX_BINDINGS,
test "$WANT_MPI_CXX_SUPPORT" = 1)
AC_DEFINE_UNQUOTED(OMPI_WANT_CXX_BINDINGS, $WANT_MPI_CXX_SUPPORT,
[Whether we want MPI cxx support or not])
##################################
# Only after setting up both
# C and C++ check compiler attributes.
##################################
ompi_show_subtitle "Compiler characteristics"
OMPI_CHECK_ATTRIBUTES
- As proposed in RFC and telcon, warn the user about deprecated functionality (per MPI-2.1). This warning can be toggled using --enable-mpi-interface-warning (default OFF), but can be selectively turned on passing mpicc -DOMPI_WANT_MPI_INTERFACE_WARNING Using icc, gcc < 4.5, warnings (such as in mpi2basic_tests) show: type_vector.c:83: warning: ‘MPI_Type_hvector’ is deprecated (declared at /home/../usr/include/mpi.h:1379) Using gcc-4.5 (gcc-svn) these show up as: type_vector.c:83: warning: ‘MPI_Type_hvector’ is deprecated (declared at /home/../usr/include/mpi.h:1379): MPI_Type_hvector is superseded by MPI_Type_create_hvector in MPI-2.0 Jeff and I propose to turn such warnings on with Open MPI-1.7 by default. - Detection of user-level compiler is handled using the preprocessor checks of GASnet's other/portable_platform.h (thanks to Paul Hargrove and Dan Bonachea) adapted into ompi/include/mpi_portable_platform.h (see comments). The OMPI-build time detection is output (Familyname and Version) with ompi_info. This functionality (actually any upcoming __attribute__) are turned off, if a different compiler (and version) is being detected. - Note, that any warnings regarding (user-compiler!=build-compiler) as discussed in the RFC are _not_ included for now. - Tested on Linux with --enable-mpi-interface-warning on Linux, gcc-4.5 (deprecated w/ specific msg) Linux, gcc-4.3 (deprecated w/o specific msg) Linux, pathscale 3.1 (deprecated w/o specific msg) Linux, icc-11.0 (deprecated w/o specific msg) Linux, PGI-8.0.6 accepts __deprecated__ but does not issue a warning, further investigation needed... This commit was SVN r21262.
2009-05-22 08:39:43 +04:00
OPAL_CHECK_COMPILER_VERSION_ID
##################################
# Assembler Configuration
##################################
ompi_show_subtitle "Assembler"
AM_PROG_AS
OMPI_CONFIG_ASM
2003-11-22 19:36:58 +03:00
##################################
# Fortran
##################################
#-----------
# Fortran 77
#-----------
OMPI_SETUP_F77
# This allows us to mark bogus types, but still have them be a valid
# [sentinel] value
AC_DEFINE([ompi_fortran_bogus_type_t], [int],
[A bogus type that allows us to have sentinel type values that are still valid])
# We want to set the #define's for all of these, so invoke the macros
# regardless of whether we have F77 support or not.
- Split the datatype engine into two parts: an MPI specific part in OMPI and a language agnostic part in OPAL. The convertor is completely moved into OPAL. This offers several benefits as described in RFC http://www.open-mpi.org/community/lists/devel/2009/07/6387.php namely: - Fewer basic types (int* and float* types, boolean and wchar - Fixing naming scheme to ompi-nomenclature. - Usability outside of the ompi-layer. - Due to the fixed nature of simple opal types, their information is completely known at compile time and therefore constified - With fewer datatypes (22), the actual sizes of bit-field types may be reduced from 64 to 32 bits, allowing reorganizing the opal_datatype structure, eliminating holes and keeping data required in convertor (upon send/recv) in one cacheline... This has implications to the convertor-datastructure and other parts of the code. - Several performance tests have been run, the netpipe latency does not change with this patch on Linux/x86-64 on the smoky cluster. - Extensive tests have been done to verify correctness (no new regressions) using: 1. mpi_test_suite on linux/x86-64 using clean ompi-trunk and ompi-ddt: a. running both trunk and ompi-ddt resulted in no differences (except for MPI_SHORT_INT and MPI_TYPE_MIX_LB_UB do now run correctly). b. with --enable-memchecker and running under valgrind (one buglet when run with static found in test-suite, commited) 2. ibm testsuite on linux/x86-64 using clean ompi-trunk and ompi-ddt: all passed (except for the dynamic/ tests failed!! as trunk/MTT) 3. compilation and usage of HDF5 tests on Jaguar using PGI and PathScale compilers. 4. compilation and usage on Scicortex. - Please note, that for the heterogeneous case, (-m32 compiled binaries/ompi), neither ompi-trunk, nor ompi-ddt branch would successfully launch. This commit was SVN r21641.
2009-07-13 08:56:31 +04:00
OMPI_F77_CHECK([CHARACTER], [yes],
[char, int32_t, int, int64_t, long long, long], [-1])
OMPI_F77_CHECK([LOGICAL], [yes],
[char, int32_t, int, int64_t, long long, long], [-1])
OMPI_F77_CHECK([LOGICAL*1], [yes],
[char, int8_t, short, int32_t, int, int64_t, long long, long], [1])
OMPI_F77_CHECK([LOGICAL*2], [yes],
[short, int16_t, int32_t, int, int64_t, long long, long], [2])
OMPI_F77_CHECK([LOGICAL*4], [yes],
[int32_t, int, int64_t, long long, long], [4])
OMPI_F77_CHECK([LOGICAL*8], [yes],
[int, int64_t, long long, long], [8])
- Split the datatype engine into two parts: an MPI specific part in OMPI and a language agnostic part in OPAL. The convertor is completely moved into OPAL. This offers several benefits as described in RFC http://www.open-mpi.org/community/lists/devel/2009/07/6387.php namely: - Fewer basic types (int* and float* types, boolean and wchar - Fixing naming scheme to ompi-nomenclature. - Usability outside of the ompi-layer. - Due to the fixed nature of simple opal types, their information is completely known at compile time and therefore constified - With fewer datatypes (22), the actual sizes of bit-field types may be reduced from 64 to 32 bits, allowing reorganizing the opal_datatype structure, eliminating holes and keeping data required in convertor (upon send/recv) in one cacheline... This has implications to the convertor-datastructure and other parts of the code. - Several performance tests have been run, the netpipe latency does not change with this patch on Linux/x86-64 on the smoky cluster. - Extensive tests have been done to verify correctness (no new regressions) using: 1. mpi_test_suite on linux/x86-64 using clean ompi-trunk and ompi-ddt: a. running both trunk and ompi-ddt resulted in no differences (except for MPI_SHORT_INT and MPI_TYPE_MIX_LB_UB do now run correctly). b. with --enable-memchecker and running under valgrind (one buglet when run with static found in test-suite, commited) 2. ibm testsuite on linux/x86-64 using clean ompi-trunk and ompi-ddt: all passed (except for the dynamic/ tests failed!! as trunk/MTT) 3. compilation and usage of HDF5 tests on Jaguar using PGI and PathScale compilers. 4. compilation and usage on Scicortex. - Please note, that for the heterogeneous case, (-m32 compiled binaries/ompi), neither ompi-trunk, nor ompi-ddt branch would successfully launch. This commit was SVN r21641.
2009-07-13 08:56:31 +04:00
OMPI_F77_CHECK([INTEGER], [yes],
[int32_t, int, int64_t, long long, long], [-1])
OMPI_F77_CHECK([INTEGER*1], [no],
[char, int8_t, short, int, int64_t, long long, long], [1])
OMPI_F77_CHECK([INTEGER*2], [no],
[short, int16_t, int32_t, int, int64_t, long long, long], [2])
OMPI_F77_CHECK([INTEGER*4], [no],
[int32_t, int, int64_t, long long, long], [4])
OMPI_F77_CHECK([INTEGER*8], [no],
[int, int64_t, long long, long], [8])
OMPI_F77_CHECK([INTEGER*16], [no],
[int, int64_t, long long, long], [16])
OMPI_F77_CHECK([REAL], [yes],
[float, double, long double], [-1])
OMPI_F77_CHECK([REAL*2], [no],
[float, double, long double], [2])
OMPI_F77_CHECK([REAL*4], [no],
[float, double, long double], [4])
OMPI_F77_CHECK([REAL*8], [no],
[float, double, long double], [8])
OMPI_F77_CHECK([REAL*16], [no],
[float, double, long double], [16])
# In some compilers, the bit representation of REAL*16 is not the same
# as the C counterpart that we found. If this is the case, then we
# want to disable reduction support for MPI_REAL16 (per ticket #1603).
OMPI_F77_CHECK_REAL16_C_EQUIV
OMPI_F77_CHECK([DOUBLE PRECISION], [yes],
[float, double, long double], [-1])
OMPI_F77_CHECK([COMPLEX], [yes], [], [-1])
# The complex*N tests are a bit different (note: the complex tests are
# the same as all the rest, because complex is a composite of two
# reals, which we *have* to have. It's only the complex*N tests that
# are different). The fortran complex types are composites of the
# real*(N/2) types. So for us to support complex*N, two conditions
# must be true:
#
# a) we must support real*(N/2) (i.e., compiler supports it and we
# have a back-end C type for it)
# b) compiler supports complex*N
OMPI_F77_CHECK([COMPLEX*8], [no], [], [8])
OMPI_F77_CHECK([COMPLEX*16], [no], [], [16])
OMPI_F77_CHECK([COMPLEX*32], [no], [], [32])
# Regardless of whether we have fortran bindings, or even a fortran
# compiler, get the max value for a fortran MPI handle (this macro
# handles the case where we don't have a fortran compiler).
OMPI_F77_GET_FORTRAN_HANDLE_MAX
#
# Check for Fortran compilers value of TRUE and for the correct assumption
# on LOGICAL for conversion into what C considers to be a true value
#
OMPI_F77_GET_VALUE_TRUE
OMPI_F77_CHECK_LOGICAL_ARRAY
#
# There are 2 layers to the MPI f77 layer. The only extra thing that
# determine f77 bindings is that fortran can be disabled by user. In
# such cases, we need to not build the target at all. One layer
# generates MPI_f77* bindings. The other layer generates PMPI_f77*
# bindings. The following conditions determine whether each (or both)
# these layers are built.
#
# Superceeding clause:
# - fortran77 bindings should be enabled, else everything is
# disabled
# 1. MPI_f77* bindings are needed if:
# - Profiling is not required
# - Profiling is required but weak symbols are not
# supported
# 2. PMPI_* bindings are needed if profiling is required. Hence we
# define 2 conditionals which tell us whether each of these layers
# need to be built or NOT
#
AM_CONDITIONAL(WANT_MPI_F77_BINDINGS_LAYER,
test \( "$WANT_MPI_PROFILING" = 0 -o "$OMPI_PROFILING_COMPILE_SEPARATELY" = 1 \) -a "$OMPI_WANT_F77_BINDINGS" = 1)
AM_CONDITIONAL(WANT_PMPI_F77_BINDINGS_LAYER,
test "$OMPI_WANT_F77_BINDINGS" = 1 -a "$WANT_MPI_PROFILING" = 1)
#-----------
# Fortran 90
#-----------
OMPI_SETUP_F90
2003-11-22 19:36:58 +03:00
# Look for the fortran module compiler flag
OMPI_F90_FIND_MODULE_INCLUDE_FLAG
# Look for all the types
OMPI_F90_IKINDS=""
OMPI_F90_RKINDS=""
OMPI_F90_CKINDS=""
- Split the datatype engine into two parts: an MPI specific part in OMPI and a language agnostic part in OPAL. The convertor is completely moved into OPAL. This offers several benefits as described in RFC http://www.open-mpi.org/community/lists/devel/2009/07/6387.php namely: - Fewer basic types (int* and float* types, boolean and wchar - Fixing naming scheme to ompi-nomenclature. - Usability outside of the ompi-layer. - Due to the fixed nature of simple opal types, their information is completely known at compile time and therefore constified - With fewer datatypes (22), the actual sizes of bit-field types may be reduced from 64 to 32 bits, allowing reorganizing the opal_datatype structure, eliminating holes and keeping data required in convertor (upon send/recv) in one cacheline... This has implications to the convertor-datastructure and other parts of the code. - Several performance tests have been run, the netpipe latency does not change with this patch on Linux/x86-64 on the smoky cluster. - Extensive tests have been done to verify correctness (no new regressions) using: 1. mpi_test_suite on linux/x86-64 using clean ompi-trunk and ompi-ddt: a. running both trunk and ompi-ddt resulted in no differences (except for MPI_SHORT_INT and MPI_TYPE_MIX_LB_UB do now run correctly). b. with --enable-memchecker and running under valgrind (one buglet when run with static found in test-suite, commited) 2. ibm testsuite on linux/x86-64 using clean ompi-trunk and ompi-ddt: all passed (except for the dynamic/ tests failed!! as trunk/MTT) 3. compilation and usage of HDF5 tests on Jaguar using PGI and PathScale compilers. 4. compilation and usage on Scicortex. - Please note, that for the heterogeneous case, (-m32 compiled binaries/ompi), neither ompi-trunk, nor ompi-ddt branch would successfully launch. This commit was SVN r21641.
2009-07-13 08:56:31 +04:00
OMPI_F90_CHECK([CHARACTER])
# LOGICAL and friends
OMPI_F90_CHECK([LOGICAL])
OMPI_F90_CHECK([LOGICAL*1], [1])
OMPI_F90_CHECK([LOGICAL*2], [2])
OMPI_F90_CHECK([LOGICAL*4], [4])
OMPI_F90_CHECK([LOGICAL*8], [8])
# INTEGER and friends
OMPI_F90_CHECK([INTEGER])
OMPI_F90_CHECK([INTEGER*1], [1])
OMPI_F90_CHECK([INTEGER*2], [2])
OMPI_F90_CHECK([INTEGER*4], [4])
OMPI_F90_CHECK([INTEGER*8], [8])
OMPI_F90_CHECK([INTEGER*16], [16])
# REAL, DOUBLE PRECISION, REAL*4, *8, *16
OMPI_F90_CHECK([REAL])
OMPI_F90_CHECK([REAL*2], [2])
OMPI_F90_CHECK([REAL*4], [4])
OMPI_F90_CHECK([REAL*8], [8])
OMPI_F90_CHECK([REAL*16], [16])
OMPI_F90_CHECK([DOUBLE PRECISION])
# COMPLEX, DOUBLE COMPLEX, COMPLEX*8, *16, *32
OMPI_F90_CHECK([COMPLEX])
OMPI_F90_CHECK([COMPLEX*8], [8])
OMPI_F90_CHECK([COMPLEX*16], [16])
OMPI_F90_CHECK([COMPLEX*32], [32])
OMPI_F90_CHECK([DOUBLE COMPLEX])
AC_SUBST(OMPI_F90_IKINDS)
AC_SUBST(OMPI_F90_RKINDS)
AC_SUBST(OMPI_F90_CKINDS)
# get kind value for Fortran MPI_INTEGER_KIND (corresponding to
# whatever is the same size as a F77 INTEGER -- for the
# most-likely-will-never-occur case where F77 INTEGER is smaller than
# an F90 INTEGER; see MPI-2 4.12.6.5
if test "$OMPI_SIZEOF_FORTRAN_INTEGER" = "2"; then
OMPI_F90_GET_INT_KIND(MPI_INTEGER_KIND, 4, OMPI_MPI_INTEGER_KIND)
elif test "$OMPI_SIZEOF_FORTRAN_INTEGER" = "4"; then
OMPI_F90_GET_INT_KIND(MPI_INTEGER_KIND, 9, OMPI_MPI_INTEGER_KIND)
elif test "$OMPI_SIZEOF_FORTRAN_INTEGER" = "8"; then
OMPI_F90_GET_INT_KIND(MPI_INTEGER_KIND, 18, OMPI_MPI_INTEGER_KIND)
elif test "$OMPI_SIZEOF_FORTRAN_INTEGER" = "16"; then
OMPI_F90_GET_INT_KIND(MPI_INTEGER_KIND, 19, OMPI_MPI_INTEGER_KIND)
AC_MSG_ERROR([Cannot support Fortran MPI_INTEGER_KIND!])
fi
AC_SUBST(OMPI_MPI_INTEGER_KIND)
# get kind value for Fortran MPI_ADDRESS_KIND (corresponding to
# whatever is big enough to hold (void*))
if test $ac_cv_sizeof_void_p = 2 ; then
OMPI_F90_GET_INT_KIND(MPI_ADDRESS_KIND, 4, OMPI_MPI_ADDRESS_KIND)
elif test $ac_cv_sizeof_void_p = 4 ; then
OMPI_F90_GET_INT_KIND(MPI_ADDRESS_KIND, 9, OMPI_MPI_ADDRESS_KIND)
elif test $ac_cv_sizeof_void_p = 8 ; then
OMPI_F90_GET_INT_KIND(MPI_ADDRESS_KIND, 18, OMPI_MPI_ADDRESS_KIND)
elif test $ac_cv_sizeof_void_p = 16 ; then
OMPI_F90_GET_INT_KIND(MPI_ADDRESS_KIND, 19, OMPI_MPI_ADDRESS_KIND)
else
AC_MSG_ERROR([Cannot support Fortran MPI_ADDRESS_KIND!])
fi
AC_SUBST(OMPI_MPI_ADDRESS_KIND)
# get kind value for Fortran MPI_OFFSET_KIND (corresponding to INTEGER*8)
OMPI_F90_GET_INT_KIND(MPI_OFFSET_KIND, 18, OMPI_MPI_OFFSET_KIND)
AC_SUBST(OMPI_MPI_OFFSET_KIND)
* Rewrite ompi_mca.m4 to use m4_defined lists of projects (ompi, orte, etc.), frameworks, and components without configure scripts instead of hard-coded shell variables (for projects and frameworks) and shell variable building (for components). * Add 3rd category of component configuration (in addition to configure scripts and no-configured components): configure.m4 components. These components can only be built as part of OMPI (like no-configure), but can provide an m4 file that is run as part of the main configure script. These macros can set whether the component should be built, along with just about any other configuration wanted. More care must be taken compared to configure components, as doing things like setting variables or calling AC_MSG_ERROR now affects the top-level configure script (so calling AC_MSG_ERROR if your component can't configure probably isn't what you want) * Added support to autogen.sh for the configure.m4-style components, as well as building up the m4_define lists ompi_mca.m4 now expects * Updated a number of macros to be more config.cache friendly (both so that config.cache can be used and so the test can be quickly run multiple times in the same configrue script): - ompi_config_asm - c_weak_symbols - c_get_alignment * Added new macros to be shared when configuring components: - ompi_objc.m4 (this actually provides AC_PROG_OBJC - don't ask...) - ompi_check_xgrid - ompi_check_tm - ompi_check_bproc * Updated a number of components to use configure.m4 instead of configure.stub - btl portals - io romio - tm ras and pls - bjs, lsf_bproc ras and bproc_seed pls - xgrid ras and pls - null iof (used by tm) This commit was SVN r6412.
2005-07-09 22:52:53 +04:00
# checkpoint results
AC_CACHE_SAVE
2003-11-22 19:36:58 +03:00
##################################
# Header files
##################################
ompi_show_title "Header file tests"
AC_CHECK_HEADERS([alloca.h aio.h arpa/inet.h dirent.h \
dlfcn.h execinfo.h err.h fcntl.h grp.h inttypes.h libgen.h \
libutil.h memory.h netdb.h netinet/in.h netinet/tcp.h \
poll.h pthread.h pty.h pwd.h sched.h stdint.h \
stdlib.h string.h strings.h stropts.h sys/fcntl.h sys/ipc.h \
sys/ioctl.h sys/mman.h sys/param.h sys/queue.h \
sys/resource.h sys/select.h sys/socket.h sys/sockio.h \
stdarg.h sys/stat.h sys/statvfs.h sys/time.h sys/tree.h \
sys/types.h sys/uio.h net/uio.h sys/utsname.h sys/wait.h syslog.h \
time.h termios.h ulimit.h unistd.h util.h utmp.h malloc.h \
ifaddrs.h sys/sysctl.h crt_externs.h regex.h \
ioLib.h sockLib.h hostLib.h shlwapi.h sys/synch.h])
# Needed to work around Darwin requiring sys/socket.h for
# net/if.h
AC_CHECK_HEADERS([net/if.h], [], [],
[#include <stdio.h>
#if STDC_HEADERS
# include <stdlib.h>
# include <stddef.h>
#else
# if HAVE_STDLIB_H
# include <stdlib.h>
# endif
#endif
#if HAVE_SYS_SOCKET_H
# include <sys/socket.h>
#endif
])
# Note that sometimes we have <stdbool.h>, but it doesn't work (e.g.,
# have both Portland and GNU installed; using pgcc will find GNU's
# <stdbool.h>, which all it does -- by standard -- is define "bool" to
# "_Bool" [see
# http://www.opengroup.org/onlinepubs/009695399/basedefs/stdbool.h.html],
# and Portland has no idea what to do with _Bool).
# So first figure out if we have <stdbool.h> (i.e., check the value of
# the macro HAVE_STDBOOL_H from the result of AC_CHECK_HEADERS,
# above). If we do have it, then check to see if it actually works.
# Define OPAL_USE_STDBOOL_H as approrpaite.
AC_CHECK_HEADERS([stdbool.h], [have_stdbool_h=1], [have_stdbool_h=0])
AC_MSG_CHECKING([if <stdbool.h> works])
if test "$have_stdbool_h" = "1"; then
AC_COMPILE_IFELSE(AC_LANG_PROGRAM([[
AC_INCLUDES_DEFAULT
#if HAVE_STDBOOL_H
#include <stdbool.h>
#endif]],
[[bool bar, foo = true; bar = foo;]]),
[OPAL_USE_STDBOOL_H=1 MSG=yes],[OPAL_USE_STDBOOL_H=0 MSG=no])
else
OPAL_USE_STDBOOL_H=0
MSG="no (don't have <stdbool.h>)"
fi
AC_DEFINE_UNQUOTED(OPAL_USE_STDBOOL_H, $OPAL_USE_STDBOOL_H,
[Whether to use <stdbool.h> or not])
AC_MSG_RESULT([$MSG])
2003-11-22 19:36:58 +03:00
* Rewrite ompi_mca.m4 to use m4_defined lists of projects (ompi, orte, etc.), frameworks, and components without configure scripts instead of hard-coded shell variables (for projects and frameworks) and shell variable building (for components). * Add 3rd category of component configuration (in addition to configure scripts and no-configured components): configure.m4 components. These components can only be built as part of OMPI (like no-configure), but can provide an m4 file that is run as part of the main configure script. These macros can set whether the component should be built, along with just about any other configuration wanted. More care must be taken compared to configure components, as doing things like setting variables or calling AC_MSG_ERROR now affects the top-level configure script (so calling AC_MSG_ERROR if your component can't configure probably isn't what you want) * Added support to autogen.sh for the configure.m4-style components, as well as building up the m4_define lists ompi_mca.m4 now expects * Updated a number of macros to be more config.cache friendly (both so that config.cache can be used and so the test can be quickly run multiple times in the same configrue script): - ompi_config_asm - c_weak_symbols - c_get_alignment * Added new macros to be shared when configuring components: - ompi_objc.m4 (this actually provides AC_PROG_OBJC - don't ask...) - ompi_check_xgrid - ompi_check_tm - ompi_check_bproc * Updated a number of components to use configure.m4 instead of configure.stub - btl portals - io romio - tm ras and pls - bjs, lsf_bproc ras and bproc_seed pls - xgrid ras and pls - null iof (used by tm) This commit was SVN r6412.
2005-07-09 22:52:53 +04:00
# checkpoint results
AC_CACHE_SAVE
2003-11-22 19:36:58 +03:00
##################################
# Types
##################################
ompi_show_title "Type tests"
# Size of pid_t
AC_CHECK_SIZEOF(pid_t)
AC_CHECK_TYPES([socklen_t, struct sockaddr_in, struct sockaddr_in6,
struct sockaddr_storage],
[], [], [AC_INCLUDES_DEFAULT
#if HAVE_SYS_SOCKET_H
#include <sys/socket.h>
#endif
#ifdef HAVE_NETINET_IN_H
#include <netinet/in.h>
#endif])
AC_CHECK_DECLS([AF_UNSPEC, PF_UNSPEC, AF_INET6, PF_INET6],
[], [], [AC_INCLUDES_DEFAULT
#if HAVE_SYS_SOCKET_H
#include <sys/socket.h>
#endif
#ifdef HAVE_NETINET_IN_H
#include <netinet/in.h>
#endif])
# SA_RESTART in signal.h
AC_MSG_CHECKING([if SA_RESTART defined in signal.h])
AC_EGREP_CPP(yes, [
#include <signal.h>
#ifdef SA_RESTART
yes
#endif ], [MSG=yes VALUE=1], [MSG=no VALUE=0])
AC_DEFINE_UNQUOTED(OPAL_HAVE_SA_RESTART, $VALUE,
[Whether we have SA_RESTART in <signal.h> or not])
AC_MSG_RESULT([$MSG])
AC_CHECK_MEMBERS([struct sockaddr.sa_len], [], [], [
#include <sys/types.h>
#if HAVE_SYS_SOCKET_H
#include <sys/socket.h>
#endif])
AC_CHECK_MEMBERS([struct dirent.d_type], [], [], [
#include <sys/types.h>
#include <dirent.h>])
AC_CHECK_MEMBERS([siginfo_t.si_fd],,,[#include <signal.h>])
AC_CHECK_MEMBERS([siginfo_t.si_band],,,[#include <signal.h>])
# checkpoint results
AC_CACHE_SAVE
2003-11-22 19:36:58 +03:00
##################################
# Libraries
##################################
ompi_show_title "Library and Function tests"
# Darwin doesn't need -lutil, as it's something other than this -lutil.
OMPI_CHECK_FUNC_LIB([openpty], [util])
AC_CHECK_LIB([nsl], [gethostbyname])
AC_CHECK_LIB([socket], [socket])
# Solaris has sched_yield in -lrt, usually in libc
OMPI_CHECK_FUNC_LIB([sched_yield], [rt])
# IRIX has dirname in -lgen, usually in libc
OMPI_CHECK_FUNC_LIB([dirname], [gen])
# Darwin doesn't need -lm, as it's a symlink to libSystem.dylib
OMPI_CHECK_FUNC_LIB([ceil], [m])
AC_CHECK_FUNCS([asprintf snprintf vasprintf vsnprintf openpty isatty getpwuid fork waitpid execve pipe ptsname setsid mmap tcgetpgrp posix_memalign strsignal sysconf syslog regcmp regexec regfree _NSGetEnviron socketpair strncpy_s _strdup usleep mkfifo])
# On some hosts, htonl is a define, so the AC_CHECK_FUNC will get
# confused. On others, it's in the standard library, but stubbed with
# the magic glibc foo as not implemented. and on other systems, it's
# just not there. This covers all cases.
AC_CACHE_CHECK([for htonl define],
[ompi_cv_htonl_define],
[AC_PREPROC_IFELSE([AC_LANG_PROGRAM([
#ifdef HAVE_SYS_TYPES_H
#include <sys/types.h>
#endif
#ifdef HAVE_NETINET_IN_H
#include <netinet/in.h>
#endif
#ifdef HAVE_ARPA_INET_H
#include <arpa/inet.h>
#endif],[
#ifndef ntohl
#error "ntohl not defined"
#endif
])], [ompi_cv_htonl_define=yes], [ompi_cv_htonl_define=no])])
AC_CHECK_FUNC([htonl], [ompi_have_htonl=yes], [ompi_have_htonl=no])
AS_IF([test "$ompi_cv_htonl_define" = "yes" -o "$ompi_have_htonl" = "yes"],
[AC_DEFINE_UNQUOTED([HAVE_UNIX_BYTESWAP], [1],
[whether unix byteswap routines -- htonl, htons, nothl, ntohs -- are available])])
#
# Make sure we can copy va_lists (need check declared, not linkable)
#
AC_CHECK_DECL(va_copy, OPAL_HAVE_VA_COPY=1, OPAL_HAVE_VA_COPY=0,
[#include <stdarg.h>])
AC_DEFINE_UNQUOTED(OPAL_HAVE_VA_COPY, $OPAL_HAVE_VA_COPY,
[Whether we have va_copy or not])
AC_CHECK_DECL(__va_copy, OPAL_HAVE_UNDERSCORE_VA_COPY=1,
OPAL_HAVE_UNDERSCORE_VA_COPY=0, [#include <stdarg.h>])
AC_DEFINE_UNQUOTED(OPAL_HAVE_UNDERSCORE_VA_COPY, $OPAL_HAVE_UNDERSCORE_VA_COPY,
[Whether we have __va_copy or not])
AC_CHECK_DECLS(__func__)
* Rewrite ompi_mca.m4 to use m4_defined lists of projects (ompi, orte, etc.), frameworks, and components without configure scripts instead of hard-coded shell variables (for projects and frameworks) and shell variable building (for components). * Add 3rd category of component configuration (in addition to configure scripts and no-configured components): configure.m4 components. These components can only be built as part of OMPI (like no-configure), but can provide an m4 file that is run as part of the main configure script. These macros can set whether the component should be built, along with just about any other configuration wanted. More care must be taken compared to configure components, as doing things like setting variables or calling AC_MSG_ERROR now affects the top-level configure script (so calling AC_MSG_ERROR if your component can't configure probably isn't what you want) * Added support to autogen.sh for the configure.m4-style components, as well as building up the m4_define lists ompi_mca.m4 now expects * Updated a number of macros to be more config.cache friendly (both so that config.cache can be used and so the test can be quickly run multiple times in the same configrue script): - ompi_config_asm - c_weak_symbols - c_get_alignment * Added new macros to be shared when configuring components: - ompi_objc.m4 (this actually provides AC_PROG_OBJC - don't ask...) - ompi_check_xgrid - ompi_check_tm - ompi_check_bproc * Updated a number of components to use configure.m4 instead of configure.stub - btl portals - io romio - tm ras and pls - bjs, lsf_bproc ras and bproc_seed pls - xgrid ras and pls - null iof (used by tm) This commit was SVN r6412.
2005-07-09 22:52:53 +04:00
# checkpoint results
AC_CACHE_SAVE
2003-11-22 19:36:58 +03:00
##################################
# System-specific tests
##################################
ompi_show_title "System-specific tests"
#
# Test to determine type of MPI_Offset. This is searched in the following order
# int64_t, long long, long, int. If none of these are 8 bytes, then we should
# search for int32_t, long long, long, int.
#
MPI_OFFSET_TYPE="not found"
MPI_OFFSET_DATATYPE="not found"
AC_MSG_CHECKING([checking for type of MPI_Offset])
if test $ac_cv_type_long_long = yes -a $ac_cv_sizeof_long_long = 8; then
MPI_OFFSET_TYPE="long long"
MPI_OFFSET_DATATYPE=MPI_LONG_LONG
MPI_OFFSET_SIZE=8
elif test $ac_cv_type_long = yes -a $ac_cv_sizeof_long = 8; then
MPI_OFFSET_TYPE="long"
MPI_OFFSET_DATATYPE=MPI_LONG
MPI_OFFSET_SIZE=8
elif test $ac_cv_sizeof_int = 8; then
MPI_OFFSET_TYPE="int"
MPI_OFFSET_DATATYPE=MPI_INT
MPI_OFFSET_SIZE=8
elif test $ac_cv_type_long_long = yes -a $ac_cv_sizeof_long_long = 4; then
MPI_OFFSET_TYPE="long long"
MPI_OFFSET_DATATYPE=MPI_LONG_LONG
MPI_OFFSET_SIZE=4
elif test $ac_cv_type_long = yes -a $ac_cv_sizeof_long = 4; then
MPI_OFFSET_TYPE="long"
MPI_OFFSET_DATATYPE=MPI_LONG
MPI_OFFSET_SIZE=4
elif test $ac_cv_sizeof_int = 4; then
MPI_OFFSET_TYPE="int"
MPI_OFFSET_DATATYPE=MPI_INT
MPI_OFFSET_SIZE=4
fi
AC_MSG_RESULT([$MPI_OFFSET_TYPE])
if test "$MPI_OFFSET_TYPE" = "not found"; then
AC_MSG_WARN([*** Unable to find the right definition for MPI_Offset])
AC_MSG_ERROR([Cannot continue])
fi
AC_DEFINE_UNQUOTED(OMPI_MPI_OFFSET_TYPE, $MPI_OFFSET_TYPE, [Type of MPI_Offset -- has to be defined here and typedef'ed later because mpi.h does not get AC SUBST's])
#
# Check for MPI_Aint type. Yes, there are platforms where
# sizeof(void*) != sizeof(long) (64 bit Windows, apparently).
#
if test $ac_cv_type_ptrdiff_t = yes ; then
opal_ptrdiff_t="ptrdiff_t"
elif test $ac_cv_sizeof_void_p -eq $ac_cv_sizeof_long ; then
opal_ptrdiff_t="long"
elif test $ac_cv_type_long_long = yes -a $ac_cv_sizeof_void_p -eq $ac_cv_sizeof_long_long ; then
opal_ptrdiff_t="long long"
else
AC_MSG_ERROR([Could not find datatype to emulate ptrdiff_t. Cannot continue])
fi
AC_DEFINE_UNQUOTED([OPAL_PTRDIFF_TYPE], [$opal_ptrdiff_t],
[type to use for ptrdiff_t])
#
# If we haven't already, figure out an MPI datatype that corresponds
# to the back-end C type of MPI_Offset.
#
AC_MSG_CHECKING([checking for an MPI datatype for MPI_Offset])
AC_MSG_RESULT([$MPI_OFFSET_DATATYPE])
if test "$MPI_OFFSET_DATATYPE" = "not found"; then
AC_MSG_WARN([*** Unable to find an MPI datatype corresponding to MPI_Offset])
AC_MSG_ERROR([Cannot continue])
fi
AC_DEFINE_UNQUOTED(OMPI_OFFSET_DATATYPE, $MPI_OFFSET_DATATYPE, [MPI datatype corresponding to MPI_Offset])
# Do we have _SC_NPROCESSORS_ONLN? (only going to pass if we also have
# <unistd.h> and sysconf(), which is ok) OS X 10.4 has <unistd.h> and
# sysconf(), but does not have _SC_NPROCESSORS_ONLN. Doh!
AC_CACHE_CHECK([for _SC_NPROCESSORS_ONLN],
[ompi_cv_have__SC_NPROCESSORS_ONLN],
[AC_COMPILE_IFELSE([AC_LANG_PROGRAM([
AC_INCLUDES_DEFAULT
#include <unistd.h>
],
[int i = _SC_NPROCESSORS_ONLN;])],
[ompi_cv_have__SC_NPROCESSORS_ONLN="yes"],
[ompi_cv_have__SC_NPROCESSORS_ONLN="no"])])
AS_IF([test "$ompi_cv_have__SC_NPROCESSORS_ONLN" = "yes"],
[result=1], [result=0])
AC_DEFINE_UNQUOTED([OPAL_HAVE__SC_NPROCESSORS_ONLN], [$result],
[Define to 1 ifyou have the declaration of _SC_NPROCESSORS_ONLN, and to 0 otherwise])
2003-11-22 19:36:58 +03:00
# all: endian
AC_C_BIGENDIAN
OMPI_CHECK_BROKEN_QSORT
AC_CACHE_CHECK([if word-sized integers must be word-size aligned],
[ompi_cv_c_word_size_align],
[AC_LANG_PUSH(C)
AC_RUN_IFELSE([AC_LANG_PROGRAM([dnl
#include <stdlib.h>], [[ long data[2] = {0, 0};
long *lp;
int *ip;
ip = (int*) data;
ip++;
lp = (long*) ip;
return lp[0]; ]])],
[ompi_cv_c_word_size_align=no],
[ompi_cv_c_word_size_align=yes],
[ompi_cv_c_word_size_align=yes])])
AS_IF([test $ompi_cv_c_word_size_align = yes], [results=1], [results=0])
AC_DEFINE_UNQUOTED([OPAL_ALIGN_WORD_SIZE_INTEGERS], [$results],
[set to 1 if word-size integers must be aligned to word-size padding to prevent bus errors])
2003-11-22 19:36:58 +03:00
# all: SYSV semaphores
# all: SYSV shared memory
# all: size of FD_SET
# all: sizeof struct stat members
# all: type of getsockopt optlen
# all: type of recvfrom optlen
#
# Check out what thread support we have
#
OMPI_CONFIG_THREADS
CFLAGS="$CFLAGS $THREAD_CFLAGS"
CXXFLAGS="$CXXFLAGS $THREAD_CXXFLAGS"
CPPFLAGS="$CPPFLAGS $THREAD_CPPFLAGS"
CXXCPPFLAGS="$CXXCPPFLAGS $THREAD_CXXCPPFLAGS"
LDFLAGS="$LDFLAGS $THREAD_LDFLAGS"
LIBS="$LIBS $THREAD_LIBS"
WRAPPER_EXTRA_CPPFLAGS="$WRAPPER_EXTRA_CPPFLAGS $THREAD_CPPFLAGS"
WRAPPER_EXTRA_CFLAGS="$WRAPPER_EXTRA_CFLAGS $THREAD_CFLAGS"
WRAPPER_EXTRA_CXXFLAGS="$WRAPPER_EXTRA_CXXFLAGS $THREAD_CXXFLAGS"
WRAPPER_EXTRA_FFLAGS="$WRAPPER_EXTRA_FFLAGS $THREAD_FFLAGS"
WRAPPER_EXTRA_FCFLAGS="$WRAPPER_EXTRA_FCFLAGS $THREAD_FFLAGS"
WRAPPER_EXTRA_LDFLAGS="$WRAPPER_EXTRA_LDFLAGS $THREAD_LDFLAGS"
# no need to update WRAPPER_EXTRA_LIBS - we'll get it from LT later
#
# What is the local equivalent of "ln -s"
#
AC_PROG_LN_S
AC_PROG_GREP
AC_PROG_EGREP
#
# We need as and lex
#
AM_PROG_AS
AM_PROG_LEX
# If we don't have GNU Flex and we don't have a generated .c file
# (distribution tarballs will have the .c file included, but SVN
# checkouts will not), then error. Must have GNU Flex -- other
# versions of Lex are not workable (all things being equal, since this
# is *only* required for developers, we decided that it really was not
# worth it to be portable between different versions of lex ;-).
if test -z "$LEX" -o -n "`echo $LEX | $GREP missing`" -o \
"`basename $LEX`" != "flex"; then
if test ! -f "$srcdir/opal/util/show_help_lex.c"; then
AC_MSG_WARN([*** Could not find GNU Flex on your system.])
AC_MSG_WARN([*** GNU Flex required for developer builds of Open MPI.])
AC_MSG_WARN([*** Other versions of Lex are not supported.])
AC_MSG_WARN([*** YOU DO NOT NEED FLEX FOR DISTRIBUTION TARBALLS!])
AC_MSG_WARN([*** If you absolutely cannot install GNU Flex on this system])
AC_MSG_WARN([*** consider using a distribution tarball, or generate the])
AC_MSG_WARN([*** following files on another system (using Flex) and])
AC_MSG_WARN([*** copy them here:])
for lfile in `find . -name \*.l -print`; do
cfile="`echo $lfile | cut -d. -f-2`"
AC_MSG_WARN([*** $cfile.c])
done
AC_MSG_ERROR([Cannot continue])
fi
fi
#
# Look for ps command and arguments for orte-clean
#
OMPI_PS_FLAVOR_CHECK
#
# File system case sensitivity
#
OMPI_CASE_SENSITIVE_FS_SETUP
2003-11-22 19:36:58 +03:00
# AIX: FIONBIO in sys/ioctl.h
# glibc: memcpy
#
# Do we have RLIMIT_NPROC in <sys/resources.h>? (e.g., Solaris does not)
#
AC_CHECK_DECLS([RLIMIT_NPROC], [], [], [
AC_INCLUDES_DEFAULT
#if HAVE_SYS_RESOURCE_H
#include <sys/resource.h>
#endif])
#
# Do we have RLIMIT_MEMLOCK in <sys/resources.h>? (e.g., Solaris does not)
#
AC_CHECK_DECLS([RLIMIT_MEMLOCK], [], [], [
AC_INCLUDES_DEFAULT
#if HAVE_SYS_RESOURCE_H
#include <sys/resource.h>
#endif])
* Rewrite ompi_mca.m4 to use m4_defined lists of projects (ompi, orte, etc.), frameworks, and components without configure scripts instead of hard-coded shell variables (for projects and frameworks) and shell variable building (for components). * Add 3rd category of component configuration (in addition to configure scripts and no-configured components): configure.m4 components. These components can only be built as part of OMPI (like no-configure), but can provide an m4 file that is run as part of the main configure script. These macros can set whether the component should be built, along with just about any other configuration wanted. More care must be taken compared to configure components, as doing things like setting variables or calling AC_MSG_ERROR now affects the top-level configure script (so calling AC_MSG_ERROR if your component can't configure probably isn't what you want) * Added support to autogen.sh for the configure.m4-style components, as well as building up the m4_define lists ompi_mca.m4 now expects * Updated a number of macros to be more config.cache friendly (both so that config.cache can be used and so the test can be quickly run multiple times in the same configrue script): - ompi_config_asm - c_weak_symbols - c_get_alignment * Added new macros to be shared when configuring components: - ompi_objc.m4 (this actually provides AC_PROG_OBJC - don't ask...) - ompi_check_xgrid - ompi_check_tm - ompi_check_bproc * Updated a number of components to use configure.m4 instead of configure.stub - btl portals - io romio - tm ras and pls - bjs, lsf_bproc ras and bproc_seed pls - xgrid ras and pls - null iof (used by tm) This commit was SVN r6412.
2005-07-09 22:52:53 +04:00
# checkpoint results
AC_CACHE_SAVE
2003-11-22 19:36:58 +03:00
##################################
# MCA
##################################
ompi_show_title "Modular Component Architecture (MCA) setup"
AC_MSG_CHECKING([for subdir args])
OMPI_CONFIG_SUBDIR_ARGS([ompi_subdir_args])
AC_MSG_RESULT([$ompi_subdir_args])
OMPI_MCA
* Rewrite ompi_mca.m4 to use m4_defined lists of projects (ompi, orte, etc.), frameworks, and components without configure scripts instead of hard-coded shell variables (for projects and frameworks) and shell variable building (for components). * Add 3rd category of component configuration (in addition to configure scripts and no-configured components): configure.m4 components. These components can only be built as part of OMPI (like no-configure), but can provide an m4 file that is run as part of the main configure script. These macros can set whether the component should be built, along with just about any other configuration wanted. More care must be taken compared to configure components, as doing things like setting variables or calling AC_MSG_ERROR now affects the top-level configure script (so calling AC_MSG_ERROR if your component can't configure probably isn't what you want) * Added support to autogen.sh for the configure.m4-style components, as well as building up the m4_define lists ompi_mca.m4 now expects * Updated a number of macros to be more config.cache friendly (both so that config.cache can be used and so the test can be quickly run multiple times in the same configrue script): - ompi_config_asm - c_weak_symbols - c_get_alignment * Added new macros to be shared when configuring components: - ompi_objc.m4 (this actually provides AC_PROG_OBJC - don't ask...) - ompi_check_xgrid - ompi_check_tm - ompi_check_bproc * Updated a number of components to use configure.m4 instead of configure.stub - btl portals - io romio - tm ras and pls - bjs, lsf_bproc ras and bproc_seed pls - xgrid ras and pls - null iof (used by tm) This commit was SVN r6412.
2005-07-09 22:52:53 +04:00
# checkpoint results
AC_CACHE_SAVE
##################################
# MPI Extended Interfaces
##################################
ompi_show_title "Extended MPI interfaces setup"
OMPI_EXT
AC_DEFINE_UNQUOTED(OMPI_EXT_COMPONENTS, "$EXT_mpiext_ALL_COMPONENTS",
[MPI Extended Interface Components])
# checkpoint results
AC_CACHE_SAVE
##################################
# Visibility
##################################
# Check the visibility declspec at the end to avoid problem with
# the previous tests that are not necessarily prepared for
# the visibility feature.
ompi_show_title "Symbol visibility feature"
OMPI_CHECK_VISIBILITY
##################################
# Contributed software
##################################
ompi_show_title "Contributed software setup"
OMPI_CONTRIB
# checkpoint results
AC_CACHE_SAVE
############################################################################
# Final top-level OMPI configuration
############################################################################
ompi_show_title "Final top-level OMPI configuration"
############################################################################
# Libevent setup
############################################################################
ompi_show_subtitle "Libevent 3rd party event handler"
CFLAGS_save="$CFLAGS"
CFLAGS="$OMPI_CFLAGS_BEFORE_PICKY"
OMPI_SETUP_LIBEVENT
CFLAGS="$CFLAGS_save"
2003-11-22 19:36:58 +03:00
############################################################################
# Libtool: part two
# (after C compiler setup = no compiler/linker tests after this)
2003-11-22 19:36:58 +03:00
############################################################################
ompi_show_subtitle "Libtool configuration"
2003-11-22 19:36:58 +03:00
# Use the undocumented solaris_use_stlport4 libtool variable to turn off any
# Cstd/stlport4 linkage. This allows Open MPI to be C++ STL agnostic.
if test "x$ompi_cv_c_compiler_vendor" = "xsun"; then
solaris_use_stlport4="yes"
fi
dnl Not all versions of LT set the PACKAGE_VERSION
m4_ifdef([LT_PACKAGE_VERSION], [], [m4_define([LT_PACKAGE_VERSION], [1.5.22])])
m4_if(m4_version_compare(m4_defn([LT_PACKAGE_VERSION]), 2.0), -1, [
AC_LIBLTDL_CONVENIENCE(opal/libltdl)
AC_LIBTOOL_DLOPEN
AC_PROG_LIBTOOL
], [
LT_CONFIG_LTDL_DIR([opal/libltdl], [subproject])
LTDL_CONVENIENCE
LT_INIT([dlopen win32-dll])
])
ompi_show_subtitle "GNU libltdl setup"
2003-11-22 19:36:58 +03:00
# AC_CONFIG_SUBDIRS appears to be broken for non-gcc compilers (i.e.,
# passing precious variables down to the sub-configure).
#
# Finally, make ltdl follow the same shared/static convention that was
# user for the main OMPI libraries. So manually examine
# $enable_shared and $enable_static and pass down the corresponding
# flags.
if test "$OPAL_ENABLE_DLOPEN_SUPPORT" = "1" ; then
ompi_subdir_args="$ompi_subdir_args --enable-ltdl-convenience --disable-ltdl-install"
if test "$enable_shared" = "yes"; then
ompi_subdir_args="$ompi_subdir_args --enable-shared"
else
ompi_subdir_args="$ompi_subdir_args --disable-shared"
fi
if test "$enable_static" = "yes"; then
ompi_subdir_args="$ompi_subdir_args --enable-static"
else
ompi_subdir_args="$ompi_subdir_args --disable-static"
fi
CFLAGS_save="$CFLAGS"
CFLAGS="$OMPI_CFLAGS_BEFORE_PICKY $OPAL_VISIBILITY_CFLAGS"
OMPI_CONFIG_SUBDIR(opal/libltdl, [$ompi_subdir_args], [HAPPY=1], [HAPPY=0])
if test "$HAPPY" = "1"; then
LIBLTDL_SUBDIR=libltdl
CPPFLAGS_save="$CPPFLAGS"
CPPFLAGS="-I."
AC_EGREP_HEADER([lt_dladvise_init], [opal/libltdl/ltdl.h],
[OPAL_HAVE_LTDL_ADVISE=1],
[OPAL_HAVE_LTDL_ADVISE=0])
CPPFLAGS="$CPPFLAGS_save"
# Arrgh. This is gross. But I can't think of any other way to do
# it. :-(
flags=`$EGREP ^LIBADD_DL opal/libltdl/Makefile | cut -d= -f2-`
OMPI_CHECK_LINKER_FLAGS([opal/libltdl/libtool], [-export-dynamic $flags])
WRAPPER_EXTRA_LIBS="$WRAPPER_EXTRA_LIBS $extra_ldflags"
LDFLAGS="-export-dynamic $LDFLAGS"
else
AC_MSG_WARN([Failed to build GNU libltdl. This usually means that something])
AC_MSG_WARN([is incorrectly setup with your environment. There may be useful information in])
AC_MSG_WARN([opal/libltdl/config.log. You can also disable GNU libltdl, which will disable])
AC_MSG_WARN([dynamic shared object loading, by configuring with --disable-dlopen.])
AC_MSG_ERROR([Cannot continue])
fi
CFLAGS="$CFLAGS_save"
else
AC_MSG_WARN([libltdl support disabled (by --disable-dlopen)])
LIBLTDL=
LIBLTDL_SUBDIR=
OPAL_HAVE_LTDL_ADVISE=0
# append instead of prepend, since LIBS are going to be system
# type things needed by everyone. Normally, libltdl will push
# these into LIBS, but since we've disabled libltdl, we need
# to do it here.
WRAPPER_EXTRA_LIBS="$WRAPPER_EXTRA_LIBS $LIBS"
fi
AC_SUBST(LTDLINCL)
AC_SUBST(LIBLTDL)
AC_SUBST(LIBLTDL_SUBDIR)
AC_DEFINE_UNQUOTED(OPAL_WANT_LIBLTDL, $OPAL_ENABLE_DLOPEN_SUPPORT,
[Whether to include support for libltdl or not])
AC_DEFINE_UNQUOTED(OPAL_HAVE_LTDL_ADVISE, $OPAL_HAVE_LTDL_ADVISE,
[Whether libltdl appears to have the lt_dladvise interface])
2003-11-22 19:36:58 +03:00
############################################################################
# final compiler config
2003-11-22 19:36:58 +03:00
############################################################################
ompi_show_subtitle "Compiler flags"
2003-11-22 19:36:58 +03:00
#
# This is needed for VPATH builds, so that it will -I the appropriate
# include directory. We delayed doing it until now just so that
2003-11-22 19:36:58 +03:00
# '-I$(top_srcdir)' doesn't show up in any of the configure output --
# purely aesthetic.
#
# Because opal_config.h, orte_config.h, and ompi_config.h are all
# created by AC_CONFIG_HEADERS, we don't need to -I the builddir for
# <project>/include. If we VPATH building, we do need to include the
# source directories, however.
#
if test "$OMPI_TOP_BUILDDIR" != "$OMPI_TOP_SRCDIR"; then
CPPFLAGS='-I$(top_srcdir) -I$(top_builddir) -I$(top_srcdir)/opal/include -I$(top_srcdir)/orte/include -I$(top_srcdir)/ompi/include'" $CPPFLAGS"
CXXCPPFLAGS='-I$(top_srcdir) -I$(top_builddir) -I$(top_srcdir)/opal/include -I$(top_srcdir)/orte/include -I$(top_srcdir)/ompi/include'" $CXXCPPFLAGS"
else
CPPFLAGS='-I$(top_srcdir)'" $CPPFLAGS"
CXXCPPFLAGS='-I$(top_srcdir)'" $CXXCPPFLAGS"
fi
2003-11-22 19:36:58 +03:00
#
# Do a final process of the CFLAGS to make a WITHOUT_OPTFLAGS version.
# We need this so that we can guarantee to build the TotalView stuff
# with -g and nothing else.
#
OMPI_MAKE_STRIPPED_FLAGS($CFLAGS)
CFLAGS_WITHOUT_OPTFLAGS="$s_result"
if test "$with_tv_debug_flags" != ""; then
TOTALVIEW_DEBUG_FLAGS="$with_tv_debug_flags"
else
# Tweak the compiler flags passed to orterun for Sun Studio SPARC
# https://svn.open-mpi.org/trac/ompi/ticket/1448
if test "x$ompi_cv_c_compiler_vendor" = "xsun" -a -n "`echo $host | $GREP sparc`"; then
TOTALVIEW_DEBUG_FLAGS="-g -xO0"
else
TOTALVIEW_DEBUG_FLAGS="-g"
fi
fi
AC_MSG_CHECKING([which of CFLAGS are ok for TotalView modules])
AC_MSG_RESULT([$CFLAGS_WITHOUT_OPTFLAGS])
AC_MSG_CHECKING([extra CFLAGS for TotalView modules])
AC_MSG_RESULT([$TOTALVIEW_DEBUG_FLAGS])
AC_SUBST(CFLAGS_WITHOUT_OPTFLAGS)
AC_SUBST(TOTALVIEW_DEBUG_FLAGS)
#
# Delayed the substitution of CFLAGS and CXXFLAGS until now because
# they may have been modified throughout the course of this script.
#
2003-11-22 19:36:58 +03:00
AC_SUBST(CFLAGS)
AC_SUBST(CPPFLAGS)
AC_SUBST(CXXFLAGS)
AC_SUBST(CXXCPPFLAGS)
AC_SUBST(FFLAGS)
AC_SUBST(FCFLAGS)
2003-11-22 19:36:58 +03:00
AC_SUBST(OMPI_LIBMPI_EXTRA_LIBS)
AC_SUBST(OMPI_LIBMPI_EXTRA_LDFLAGS)
#
# Aggregate MCA parameters directory
#
AC_SUBST([AMCA_PARAM_SETS_DIR], ['$(pkgdatadir)/amca-param-sets'])
############################################################################
# final wrapper compiler config
############################################################################
ompi_show_subtitle "Wrapper compiler flags"
OMPI_SETUP_WRAPPER_FINAL
2003-11-22 19:36:58 +03:00
# Recreate some defines prefixed with OMPI_ so that there are no bare
# autoconf macro defines in mpi.h. Since AC sometimes changes whether
# things are defined as null tokens or an integer result, two projects
# with different versions of AC can cause problems.
if test $ac_cv_header_stdc = yes; then
AC_DEFINE(OPAL_STDC_HEADERS, 1,
[Do not use outside of mpi.h. Define to 1 if you have the ANSI C header files.])
fi
if test $ac_cv_header_sys_time_h = yes ; then
AC_DEFINE(OPAL_HAVE_SYS_TIME_H, 1,
[Do not use outside of mpi.h. Define to 1 if you have the <sys/time.h> header file.])
fi
if test $ac_cv_type_long_long = yes ; then
AC_DEFINE(OPAL_HAVE_LONG_LONG, 1,
[Do not use outside of mpi.h. Define to 1 if the system has the type `long long'.]) dnl `
fi
if test $ac_cv_header_sys_synch_h = yes ; then
AC_DEFINE(OPAL_HAVE_SYS_SYNCH_H, 1,
[Do not use outside of mpi.h. Define to 1 if you have the <sys/synch.h> header file.])
fi
2003-11-22 19:36:58 +03:00
############################################################################
# Party on
############################################################################
ompi_show_subtitle "Final output"
2003-11-22 19:36:58 +03:00
AC_CONFIG_FILES([
Makefile
config/Makefile
contrib/Makefile
opal/Makefile
opal/etc/Makefile
opal/include/Makefile
opal/asm/Makefile
- Split the datatype engine into two parts: an MPI specific part in OMPI and a language agnostic part in OPAL. The convertor is completely moved into OPAL. This offers several benefits as described in RFC http://www.open-mpi.org/community/lists/devel/2009/07/6387.php namely: - Fewer basic types (int* and float* types, boolean and wchar - Fixing naming scheme to ompi-nomenclature. - Usability outside of the ompi-layer. - Due to the fixed nature of simple opal types, their information is completely known at compile time and therefore constified - With fewer datatypes (22), the actual sizes of bit-field types may be reduced from 64 to 32 bits, allowing reorganizing the opal_datatype structure, eliminating holes and keeping data required in convertor (upon send/recv) in one cacheline... This has implications to the convertor-datastructure and other parts of the code. - Several performance tests have been run, the netpipe latency does not change with this patch on Linux/x86-64 on the smoky cluster. - Extensive tests have been done to verify correctness (no new regressions) using: 1. mpi_test_suite on linux/x86-64 using clean ompi-trunk and ompi-ddt: a. running both trunk and ompi-ddt resulted in no differences (except for MPI_SHORT_INT and MPI_TYPE_MIX_LB_UB do now run correctly). b. with --enable-memchecker and running under valgrind (one buglet when run with static found in test-suite, commited) 2. ibm testsuite on linux/x86-64 using clean ompi-trunk and ompi-ddt: all passed (except for the dynamic/ tests failed!! as trunk/MTT) 3. compilation and usage of HDF5 tests on Jaguar using PGI and PathScale compilers. 4. compilation and usage on Scicortex. - Please note, that for the heterogeneous case, (-m32 compiled binaries/ompi), neither ompi-trunk, nor ompi-ddt branch would successfully launch. This commit was SVN r21641.
2009-07-13 08:56:31 +04:00
opal/datatype/Makefile
opal/event/Makefile
opal/event/compat/Makefile
opal/event/compat/sys/Makefile
opal/util/Makefile
opal/util/keyval/Makefile
opal/mca/base/Makefile
opal/tools/wrappers/Makefile
opal/tools/wrappers/opalcc-wrapper-data.txt
opal/tools/wrappers/opalc++-wrapper-data.txt
opal/tools/opal-checkpoint/Makefile
opal/tools/opal-restart/Makefile
orte/Makefile
orte/include/Makefile
orte/etc/Makefile
orte/tools/orted/Makefile
orte/tools/orterun/Makefile
orte/tools/wrappers/Makefile
orte/tools/wrappers/ortecc-wrapper-data.txt
orte/tools/wrappers/ortec++-wrapper-data.txt
orte/tools/orte-checkpoint/Makefile
orte/tools/orte-iof/Makefile
orte/tools/orte-restart/Makefile
orte/tools/orte-ps/Makefile
orte/tools/orte-clean/Makefile
orte/tools/orte-top/Makefile
orte/tools/orte-bootproxy/Makefile
ompi/Makefile
ompi/etc/Makefile
ompi/include/Makefile
ompi/include/mpif.h
ompi/include/mpif-config.h
ompi/datatype/Makefile
ompi/debuggers/Makefile
ompi/mpi/c/Makefile
ompi/mpi/c/profile/Makefile
ompi/mpi/cxx/Makefile
ompi/mpi/f77/Makefile
ompi/mpi/f77/profile/Makefile
ompi/mpi/f90/Makefile
ompi/mpi/f90/fortran_kinds.sh
ompi/mpi/f90/fortran_sizes.h
ompi/mpi/f90/scripts/Makefile
ompi/tools/ompi_info/Makefile
ompi/tools/wrappers/Makefile
ompi/tools/wrappers/mpicc-wrapper-data.txt
ompi/tools/wrappers/mpic++-wrapper-data.txt
ompi/tools/wrappers/mpif77-wrapper-data.txt
ompi/tools/wrappers/mpif90-wrapper-data.txt
ompi/tools/ortetools/Makefile
ompi/tools/ompi-server/Makefile
Enable modex-less launch. Consists of: 1. minor modification to include two new opal MCA params: (a) opal_profile: outputs what components were selected by each framework currently enabled for most, but not all, frameworks (b) opal_profile_file: name of file that contains profile info required for modex 2. introduction of two new tools: (a) ompi-probe: MPI process that simply calls MPI_Init/Finalize with opal_profile set. Also reports back the rml IP address for all interfaces on the node (b) ompi-profiler: uses ompi-probe to create the profile_file, also reports out a summary of what framework components are actually being used to help with configuration options 3. modification of the grpcomm basic component to utilize the profile file in place of the modex where possible 4. modification of orterun so it properly sees opal mca params and handles opal_profile correctly to ensure we don't get its profile 5. similar mod to orted as for orterun 6. addition of new test that calls orte_init followed by calls to grpcomm.barrier This is all completely benign unless actively selected. At the moment, it only supports modex-less launch for openib-based systems. Minor mod to the TCP btl would be required to enable it as well, if people are interested. Similarly, anyone interested in enabling other BTL's for modex-less operation should let me know and I'll give you the magic details. This seems to significantly improve scalability provided the file can be locally located on the nodes. I'm looking at an alternative means of disseminating the info (perhaps in launch message) as an option for removing that constraint. This commit was SVN r20098.
2008-12-10 02:49:02 +03:00
ompi/tools/ompi-probe/Makefile
ompi/tools/ompi-profiler/Makefile
test/Makefile
test/event/Makefile
test/asm/Makefile
test/class/Makefile
test/support/Makefile
test/threads/Makefile
test/peruse/Makefile
test/datatype/Makefile
2003-11-22 19:36:58 +03:00
])
AC_OUTPUT