ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность

Merge pull request #7174 from jsquyres/pr/ofi-mtl-fi-version-bump

Просмотр исходного кода 
mtl/ofi: increase the FI_VERSION requested to 1.5 and make sure to check for OFI_LOCAL_COMM
Этот коммит содержится в:
Jeff Squyres 2020-01-13 13:22:22 -05:00 коммит произвёл GitHub
родитель dd2d7d2866 49128a7adb
Коммит 887400c878
Не найден ключ, соответствующий данной подписи
Идентификатор ключа GPG: 4AEE18F83AFDEB23
4 изменённых файлов: 149 добавлений и 43 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

41
config/opal_check_ofi.m4
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 --------------------------------------------------------

8
ompi/mca/mtl/ofi/configure.m4
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])

119
ompi/mca/mtl/ofi/mtl_ofi_component.c
Просмотреть файл

@ -341,21 +341,12 @@ is_in_list(char **list, char *item)
}
static struct fi_info*
select_ofi_provider(struct fi_info *providers)
select_ofi_provider(struct fi_info *providers,
char **include_list, char **exclude_list)
{
char **include_list = NULL;
char **exclude_list = NULL;
struct fi_info *prov = providers;
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: mtl:ofi:provider_include = \"%s\"\n",
__FILE__, __LINE__, prov_include);
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: mtl:ofi:provider_exclude = \"%s\"\n",
__FILE__, __LINE__, prov_exclude);
if (NULL != prov_include) {
include_list = opal_argv_split(prov_include, ',');
if (NULL != include_list) {
while ((NULL != prov) &&
(!is_in_list(include_list, prov->fabric_attr->prov_name))) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
@ -364,8 +355,7 @@ select_ofi_provider(struct fi_info *providers)
prov->fabric_attr->prov_name);
prov = prov->next;
}
} else if (NULL != prov_exclude) {
exclude_list = opal_argv_split(prov_exclude, ',');
} else if (NULL != exclude_list) {
while ((NULL != prov) &&
(is_in_list(exclude_list, prov->fabric_attr->prov_name))) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
@ -376,9 +366,6 @@ select_ofi_provider(struct fi_info *providers)
}
}
opal_argv_free(include_list);
opal_argv_free(exclude_list);
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: mtl:ofi:prov: %s\n",
__FILE__, __LINE__,
@ -621,7 +608,9 @@ ompi_mtl_ofi_component_init(bool enable_progress_threads,
int ret, fi_version;
int num_local_ranks, sep_support_in_provider, max_ofi_ctxts;
int ofi_tag_leading_zeros, ofi_tag_bits_for_cid;
struct fi_info *hints;
char **include_list = NULL;
char **exclude_list = NULL;
struct fi_info *hints, *hints_dup = NULL;
struct fi_info *providers = NULL;
struct fi_info *prov = NULL;
struct fi_info *prov_cq_data = NULL;
@ -630,6 +619,19 @@ ompi_mtl_ofi_component_init(bool enable_progress_threads,
int universe_size;
char *univ_size_str;
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: mtl:ofi:provider_include = \"%s\"\n",
__FILE__, __LINE__, prov_include);
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: mtl:ofi:provider_exclude = \"%s\"\n",
__FILE__, __LINE__, prov_exclude);
if (NULL != prov_include) {
include_list = opal_argv_split(prov_include, ',');
} else if (NULL != prov_exclude) {
exclude_list = opal_argv_split(prov_exclude, ',');
}
/**
* Hints to filter providers
* See man fi_getinfo for a list of all filters
@ -647,9 +649,11 @@ 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 +701,59 @@ 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);
/**
* The EFA provider in Libfabric versions prior to 1.10 contains a bug
* where the FI_LOCAL_COMM and FI_REMOTE_COMM capabilities are not
* advertised. However, we know that this provider supports both local and
* remote communication. We must exclude these capability bits in order to
* select EFA when we are using a version of Libfabric with this bug.
*
* Call fi_getinfo() without those capabilities and specifically ask for
* the EFA provider. This is safe to do as EFA is only supported on Amazon
* EC2 and EC2 only supports EFA and TCP-based networks. We'll also skip
* this logic if the user specifies an include list without EFA or adds EFA
* to the exclude list.
*/
if ((include_list && is_in_list(include_list, "efa")) ||
(exclude_list && !is_in_list(exclude_list, "efa"))) {
hints_dup = fi_dupinfo(hints);
hints_dup->caps &= ~(FI_LOCAL_COMM | FI_REMOTE_COMM);
hints_dup->fabric_attr->prov_name = strdup("efa");
ret = fi_getinfo(fi_version, NULL, NULL, 0ULL, hints_dup, &providers);
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: EFA specific fi_getinfo(): %s\n",
__FILE__, __LINE__, fi_strerror(-ret));
if (FI_ENODATA == -ret) {
/**
* EFA is not available so fall through to call fi_getinfo() again
* with the local/remote capabilities set.
*/
fi_freeinfo(hints_dup);
hints_dup = NULL;
} else if (0 != ret) {
opal_show_help("help-mtl-ofi.txt", "OFI call fail", true,
"fi_getinfo",
ompi_process_info.nodename, __FILE__, __LINE__,
fi_strerror(-ret), -ret);
goto error;
} else {
fi_freeinfo(hints);
hints = hints_dup;
hints_dup = NULL;
goto select_prov;
}
}
/**
* fi_getinfo: returns information about fabric services for reaching a
@ -711,6 +766,11 @@ ompi_mtl_ofi_component_init(bool enable_progress_threads,
0ULL, /* Optional flag */
hints, /* In: Hints to filter providers */
&providers); /* Out: List of matching providers */
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: fi_getinfo(): %s\n",
__FILE__, __LINE__, fi_strerror(-ret));
if (FI_ENODATA == -ret) {
// It is not an error if no information is returned.
goto error;
@ -722,10 +782,11 @@ ompi_mtl_ofi_component_init(bool enable_progress_threads,
goto error;
}
select_prov:
/**
* Select a provider from the list returned by fi_getinfo().
*/
prov = select_ofi_provider(providers);
prov = select_ofi_provider(providers, include_list, exclude_list);
if (!prov) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: select_ofi_provider: no provider found\n",
@ -733,6 +794,11 @@ ompi_mtl_ofi_component_init(bool enable_progress_threads,
goto error;
}
opal_argv_free(include_list);
include_list = NULL;
opal_argv_free(exclude_list);
exclude_list = NULL;
/**
* Select the format of the OFI tag
*/
@ -1006,6 +1072,12 @@ ompi_mtl_ofi_component_init(bool enable_progress_threads,
return &ompi_mtl_ofi.base;
error:
if (include_list) {
opal_argv_free(include_list);
}
if (exclude_list) {
opal_argv_free(exclude_list);
}
if (providers) {
(void) fi_freeinfo(providers);
}
@ -1015,6 +1087,9 @@ error:
if (hints) {
(void) fi_freeinfo(hints);
}
if (hints_dup) {
(void) fi_freeinfo(hints_dup);
}
if (ompi_mtl_ofi.sep) {
(void) fi_close((fid_t)ompi_mtl_ofi.sep);
}

24
opal/mca/btl/usnic/configure.m4
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" ],