1
1

mtl/ofi: check for FI_LOCAL_COMM+FI_REMOTE_COMM

Make sure to get an RDM provider that can provide both local and
remote communication.  We need this check because some providers could
be selected via RXD or RXM, but can't provide local communication, for
example.

Add OPAL_CHECK_OFI_VERSION_GE() m4 macro to check that the Libfabric
we're building against is >= a target version.  Use this check in two
places:

1. MTL/OFI: Make sure it is >= v1.5, because the FI_LOCAL_COMM /
   FI_REMOTE_COMM constants were introduced in Libfabric API v1.5.
2. BTL/usnic: It already had similar configury to check for Libfabric
   >= v1.1, but the usnic component was checking for >= v1.3.  So
   update the btl/usnic configury to use the new macro and check for
   >= v1.3.

Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
Этот коммит содержится в:
Jeff Squyres 2019-11-18 07:29:39 -08:00
родитель dd2d7d2866
Коммит 21bc9042e1
4 изменённых файлов: 62 добавлений и 24 удалений

Просмотреть файл

@ -1,6 +1,6 @@
dnl -*- shell-script -*-
dnl
dnl Copyright (c) 2015-2019 Cisco Systems, Inc. All rights reserved.
dnl Copyright (c) 2015-2020 Cisco Systems, Inc. All rights reserved.
dnl Copyright (c) 2016-2017 Los Alamos National Security, LLC. All rights
dnl reserved.
dnl $COPYRIGHT$
@ -10,6 +10,45 @@ dnl
dnl $HEADER$
dnl
dnl
dnl OPAL_CHECK_OFI_VERSION_GE
dnl
dnl Check that the OFI API version number is >= a specific value.
dnl
dnl $1: version number to compare, in the form of "major,minor"
dnl (without quotes) -- i.e., a single token representing the
dnl arguments to FI_VERSION()
dnl $2: action if OFI API version is >= $1
dnl $3: action if OFI API version is < $1
AC_DEFUN([OPAL_CHECK_OFI_VERSION_GE],[
OPAL_VAR_SCOPE_PUSH([opal_ofi_ver_ge_save_CPPFLAGS opal_ofi_ver_ge_happy])
AC_MSG_CHECKING([if OFI API version number is >= $1])
opal_ofi_ver_ge_save_CPPFLAGS=$CPPFLAGS
CPPFLAGS=$opal_ofi_CPPFLAGS
AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[#include <rdma/fabric.h>]],
[[
#if !defined(FI_MAJOR_VERSION)
#error "we cannot check the version -- sad panda"
#elif FI_VERSION_LT(FI_VERSION(FI_MAJOR_VERSION, FI_MINOR_VERSION), FI_VERSION($1))
#error "version is too low -- nopes"
#endif
]])],
[opal_ofi_ver_ge_happy=1],
[opal_ofi_ver_ge_happy=0])
AS_IF([test $opal_ofi_ver_ge_happy -eq 1],
[AC_MSG_RESULT([yes])
$2],
[AC_MSG_RESULT([no])
$3])
CPPFLAGS=$opal_ofi_ver_ge_save_CPPFLAGS
OPAL_VAR_SCOPE_POP
])dnl
dnl
dnl _OPAL_CHECK_OFI
dnl --------------------------------------------------------

Просмотреть файл

@ -2,7 +2,7 @@
#
# Copyright (c) 2013-2014 Intel, Inc. All rights reserved
#
# Copyright (c) 2014-2019 Cisco Systems, Inc. All rights reserved
# Copyright (c) 2014-2020 Cisco Systems, Inc. All rights reserved
# Copyright (c) 2017 Los Alamos National Security, LLC. All rights
# reserved.
# $COPYRIGHT$
@ -28,6 +28,12 @@ AC_DEFUN([MCA_ompi_mtl_ofi_CONFIG],[
# Check for OFI
OPAL_CHECK_OFI
# The OFI MTL requires at least OFI libfabric v1.5.
AS_IF([test "$opal_ofi_happy" = "yes"],
[OPAL_CHECK_OFI_VERSION_GE([1,5],
[],
[opal_ofi_happy=no])])
AS_IF([test "$opal_ofi_happy" = "yes"],
[$1],
[$2])

Просмотреть файл

@ -647,9 +647,11 @@ ompi_mtl_ofi_component_init(bool enable_progress_threads,
__FILE__, __LINE__);
goto error;
}
/* Make sure to get a RDM provider that can do the tagged matching
interface and local communication and remote communication. */
hints->mode = FI_CONTEXT;
hints->ep_attr->type = FI_EP_RDM; /* Reliable datagram */
hints->caps = FI_TAGGED; /* Tag matching interface */
hints->ep_attr->type = FI_EP_RDM;
hints->caps = FI_TAGGED | FI_LOCAL_COMM | FI_REMOTE_COMM;
hints->tx_attr->msg_order = FI_ORDER_SAS;
hints->rx_attr->msg_order = FI_ORDER_SAS;
hints->rx_attr->op_flags = FI_COMPLETION;
@ -697,8 +699,13 @@ ompi_mtl_ofi_component_init(bool enable_progress_threads,
* FI_VERSION provides binary backward and forward compatibility support
* Specify the version of OFI is coded to, the provider will select struct
* layouts that are compatible with this version.
*
* Note: API version 1.5 is the first version that supports
* FI_LOCAL_COMM / FI_REMOTE_COMM checking (and we definitely need
* that checking -- e.g., some providers are suitable for RXD or
* RXM, but can't provide local communication).
*/
fi_version = FI_VERSION(1, 0);
fi_version = FI_VERSION(1, 5);
/**
* fi_getinfo: returns information about fabric services for reaching a

Просмотреть файл

@ -12,7 +12,7 @@
# All rights reserved.
# Copyright (c) 2006 Sandia National Laboratories. All rights
# reserved.
# Copyright (c) 2010-2019 Cisco Systems, Inc. All rights reserved
# Copyright (c) 2010-2020 Cisco Systems, Inc. All rights reserved
# Copyright (c) 2017 Los Alamos National Security, LLC. All rights
# reserved.
# $COPYRIGHT$
@ -100,25 +100,11 @@ AC_DEFUN([_OPAL_BTL_USNIC_DO_CONFIG],[
OPAL_CHECK_OFI
opal_btl_usnic_happy=$opal_ofi_happy])
# The usnic BTL requires at least OFI libfabric v1.1 (there was a
# critical bug in libfabric v1.0).
# The usnic BTL requires at least OFI libfabric v1.3.
AS_IF([test "$opal_btl_usnic_happy" = "yes"],
[AC_MSG_CHECKING([whether OFI libfabric is >= v1.1])
opal_btl_usnic_CPPFLAGS_save=$CPPFLAGS
CPPFLAGS="$opal_ofi_CPPFLAGS $CPPFLAGS"
AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[#include <rdma/fabric.h>]],
[[
#if !defined(FI_MAJOR_VERSION)
#error your version of OFI libfabric is too old
#elif FI_VERSION(FI_MAJOR_VERSION, FI_MINOR_VERSION) < FI_VERSION(1, 1)
#error your version of OFI libfabric is too old
#endif
]])],
[opal_btl_usnic_happy=yes],
[opal_btl_usnic_happy=no])
AC_MSG_RESULT([$opal_btl_usnic_happy])
CPPFLAGS=$opal_btl_usnic_CPPFLAGS_save
])
[OPAL_CHECK_OFI_VERSION_GE([1,3],
[],
[opal_btl_usnic_happy=no])])
# Make sure we can find the OFI libfabric usnic extensions header
AS_IF([test "$opal_btl_usnic_happy" = "yes" ],