Skip to content

Commit 4bc44b5

Browse files
samcmillsamcmill
authored
Update NCCL building block and bump default version to 2.29.7-1 (#536)
1 parent 8cc2d49 commit 4bc44b5

3 files changed

Lines changed: 117 additions & 80 deletions

File tree

‎docs/building_blocks.md‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -2958,7 +2958,7 @@ source. The default value is False.
29582958
use the latest commit on the default branch for the repository.
29592959

29602960
- __cuda__: Flag to specify the CUDA version of the package to download.
2961-
The default is `11.6`. This option is ignored if build is True.
2961+
The default is `13.2`. This option is ignored if build is True.
29622962

29632963
- __environment__: Boolean flag to specify whether the environment
29642964
(`CPATH`, `LD_LIBRARY_PATH`, `LIBRARY_PATH`, and `PATH`) should be
@@ -2980,7 +2980,7 @@ repository. The default is empty, i.e., use the release package
29802980
specified by `version`.
29812981

29822982
- __version__: The version of NCCL to install. The default value is
2983-
`2.12.10-1`.
2983+
`2.29.7-1`.
29842984

29852985
__Examples__
29862986

‎hpccm/building_blocks/nccl.py‎

Lines changed: 57 additions & 20 deletions
Original file line numberDiff line numberDiff line change
@@ -31,8 +31,7 @@
3131
from hpccm.building_blocks.base import bb_base
3232
from hpccm.building_blocks.generic_build import generic_build
3333
from hpccm.building_blocks.packages import packages
34-
from hpccm.common import linux_distro
35-
from hpccm.config import get_cpu_architecture
34+
from hpccm.common import cpu_arch, linux_distro
3635
from hpccm.primitives.comment import comment
3736
from hpccm.primitives.copy import copy
3837
from hpccm.primitives.environment import environment
@@ -56,7 +55,7 @@ class nccl(bb_base, hpccm.templates.downloader, hpccm.templates.envvars,
5655
use the latest commit on the default branch for the repository.
5756
5857
cuda: Flag to specify the CUDA version of the package to download.
59-
The default is `11.6`. This option is ignored if build is True.
58+
The default is `13.2`. This option is ignored if build is True.
6059
6160
environment: Boolean flag to specify whether the environment
6261
(`CPATH`, `LD_LIBRARY_PATH`, `LIBRARY_PATH`, and `PATH`) should be
@@ -78,7 +77,7 @@ class nccl(bb_base, hpccm.templates.downloader, hpccm.templates.envvars,
7877
specified by `version`.
7978
8079
version: The version of NCCL to install. The default value is
81-
`2.12.10-1`.
80+
`2.29.7-1`.
8281
8382
# Examples
8483
@@ -97,39 +96,47 @@ def __init__(self, **kwargs):
9796

9897
super(nccl, self).__init__(**kwargs)
9998

99+
self.__arch_label = '' # Filled in by __cpu_arch
100100
self.__baseurl = kwargs.pop('baseurl', 'https://github.com/NVIDIA/nccl/archive')
101101
self.__build = kwargs.pop('build', False)
102102
self.__build_environment = '' # Filled in by __configure
103103
self.__default_repository = 'https://github.com/NVIDIA/nccl.git'
104104
self.__distro_label = '' # Filled in by __distro
105-
self.__cuda = kwargs.pop('cuda', '11.6')
105+
self.__cuda = kwargs.pop('cuda', '13.2')
106106
self.__make_variables = kwargs.pop('make_variables', {})
107107
self.__ospackages = kwargs.pop('ospackages', [])
108108
self.__prefix = kwargs.pop('prefix', '/usr/local/nccl')
109+
self.__repo_key = '' # Filled in by __repo_key
109110
self.__src_directory = kwargs.pop('src_directory', None)
110-
self.__version = kwargs.pop('version', '2.12.10-1')
111+
self.__version = kwargs.pop('version', '2.29.7-1')
111112
self.__wd = kwargs.get('wd', hpccm.config.g_wd) # working directory
112113

113114
if not self.__build:
114115
# Install prebuild package
115116

117+
# Set the CPU architecture specific parameters
118+
self.__cpu_arch()
119+
116120
# Set the Linux distribution specific parameters
117121
self.__distro()
118122

123+
# Set the repo key
124+
self.__repo()
125+
119126
self += comment('NCCL {}'.format(self.__version))
120127
self += packages(ospackages=self.__ospackages)
121128
self += packages(
122129
apt=['libnccl2={0}+cuda{1}'.format(self.__version,
123130
self.__cuda),
124131
'libnccl-dev={0}+cuda{1}'.format(self.__version,
125132
self.__cuda)],
126-
apt_keys=['https://developer.download.nvidia.com/compute/cuda/repos/{0}/{1}/3bf863cc.pub'.format(self.__distro_label, get_cpu_architecture())],
127-
apt_repositories=['deb [signed-by=/usr/share/keyrings/3bf863cc.gpg] https://developer.download.nvidia.com/compute/cuda/repos/{0}/{1} /'.format(self.__distro_label, get_cpu_architecture())],
133+
apt_keys=['https://developer.download.nvidia.com/compute/cuda/repos/{0}/{1}/{2}'.format(self.__distro_label, self.__arch_label, self.__repo_key)],
134+
apt_repositories=['deb [signed-by=/usr/share/keyrings/{2}] https://developer.download.nvidia.com/compute/cuda/repos/{0}/{1} /'.format(self.__distro_label, self.__arch_label, self.__repo_key.replace('.pub', '.gpg'))],
128135
yum=['libnccl-{0}+cuda{1}'.format(self.__version, self.__cuda),
129136
'libnccl-devel-{0}+cuda{1}'.format(self.__version,
130137
self.__cuda)],
131-
yum_keys=['https://developer.download.nvidia.com/compute/cuda/repos/{0}/{1}/3bf863cc.pub'.format(self.__distro_label, get_cpu_architecture())],
132-
yum_repositories=['https://developer.download.nvidia.com/compute/cuda/repos/{0}/{1}'.format(self.__distro_label, get_cpu_architecture())])
138+
yum_keys=['https://developer.download.nvidia.com/compute/cuda/repos/{0}/{1}/{2}'.format(self.__distro_label, self.__arch_label, self.__repo_key)],
139+
yum_repositories=['https://developer.download.nvidia.com/compute/cuda/repos/{0}/{1}'.format(self.__distro_label, self.__arch_label)])
133140

134141
else:
135142
# Build from source
@@ -166,6 +173,17 @@ def __init__(self, **kwargs):
166173
self += packages(ospackages=self.__ospackages)
167174
self += self.__bb
168175

176+
def __cpu_arch(self):
177+
"""Based on the CPU architecture, set values accordingly. A user
178+
specified value overrides any defaults."""
179+
180+
if hpccm.config.g_cpu_arch == cpu_arch.AARCH64:
181+
self.__arch_label = 'sbsa'
182+
elif hpccm.config.g_cpu_arch == cpu_arch.X86_64:
183+
self.__arch_label = 'x86_64'
184+
else: # pragma: no cover
185+
raise RuntimeError('Unknown CPU architecture')
186+
169187
def __configure(self):
170188
"""Setup build options based on user parameters"""
171189

@@ -192,16 +210,20 @@ def __distro(self):
192210
self.__ospackages = ['apt-transport-https', 'ca-certificates',
193211
'gnupg', 'wget']
194212

195-
if hpccm.config.g_linux_version >= Version('18.0'):
196-
self.__distro_label = 'ubuntu1804'
213+
if hpccm.config.g_linux_version >= Version('24.0'):
214+
self.__distro_label = 'ubuntu2404'
215+
elif hpccm.config.g_linux_version >= Version('22.0'):
216+
self.__distro_label = 'ubuntu2204'
197217
else:
198-
self.__distro_label = 'ubuntu1604'
218+
self.__distro_label = 'ubuntu2004'
199219

200220
elif hpccm.config.g_linux_distro == linux_distro.CENTOS:
201-
if hpccm.config.g_linux_version >= Version('8.0'):
202-
self.__distro_label = 'rhel8'
221+
if hpccm.config.g_linux_version >= Version('10.0'):
222+
self.__distro_label = 'rhel10'
223+
elif hpccm.config.g_linux_version >= Version('9.0'):
224+
self.__distro_label = 'rhel9'
203225
else:
204-
self.__distro_label = 'rhel7'
226+
self.__distro_label = 'rhel8'
205227

206228
else: # pragma: no cover
207229
raise RuntimeError('Unknown Linux distribution')
@@ -225,6 +247,21 @@ def __download(self):
225247
if not self.repository and not self.url:
226248
self.url = '{0}/v{1}.tar.gz'.format(self.__baseurl, self.__version)
227249

250+
def __repo(self):
251+
"""Based on the Linux distribution and CPU architecture, set values
252+
accordingly. A user specified value overrides any defaults.
253+
"""
254+
255+
if self.__distro_label.startswith('ubuntu'):
256+
self.__repo_key = '3bf863cc.pub'
257+
elif self.__distro_label.startswith('rhel'):
258+
if hpccm.config.g_linux_version >= Version('10.0'):
259+
self.__repo_key = 'CDF6BA43.pub'
260+
else:
261+
self.__repo_key = 'D42D0685.pub'
262+
else: # pragma: no cover
263+
raise RuntimeError('Unknown repository')
264+
228265
def runtime(self, _from='0'):
229266
"""Generate the set of instructions to install the runtime specific
230267
components from a build in a previous stage.
@@ -246,10 +283,10 @@ def runtime(self, _from='0'):
246283
self.rt += packages(
247284
apt=['libnccl2={0}+cuda{1}'.format(self.__version,
248285
self.__cuda)],
249-
apt_keys=['https://developer.download.nvidia.com/compute/cuda/repos/{0}/{1}/3bf863cc.pub'.format(self.__distro_label, get_cpu_architecture())],
250-
apt_repositories=['deb [signed-by=/usr/share/keyrings/3bf863cc.gpg] https://developer.download.nvidia.com/compute/cuda/repos/{0}/{1} /'.format(self.__distro_label, get_cpu_architecture())],
286+
apt_keys=['https://developer.download.nvidia.com/compute/cuda/repos/{0}/{1}/{2}'.format(self.__distro_label, self.__arch_label, self.__repo_key)],
287+
apt_repositories=['deb [signed-by=/usr/share/keyrings/{2}] https://developer.download.nvidia.com/compute/cuda/repos/{0}/{1} /'.format(self.__distro_label, self.__arch_label, self.__repo_key.replace('.pub', '.gpg'))],
251288
yum=['libnccl-{0}+cuda{1}'.format(self.__version, self.__cuda)],
252-
yum_keys=['https://developer.download.nvidia.com/compute/cuda/repos/{0}/{1}/3bf863cc.pub'.format(self.__distro_label, get_cpu_architecture())],
253-
yum_repositories=['https://developer.download.nvidia.com/compute/cuda/repos/{0}/{1}'.format(self.__distro_label, get_cpu_architecture())])
289+
yum_keys=['https://developer.download.nvidia.com/compute/cuda/repos/{0}/{1}/{2}'.format(self.__distro_label, self.__arch_label, self.__repo_key)],
290+
yum_repositories=['https://developer.download.nvidia.com/compute/cuda/repos/{0}/{1}'.format(self.__distro_label, self.__arch_label)])
254291

255292
return str(self.rt)

0 commit comments

Comments
 (0)