[aws-ofi-nccl] [package list]

aws-ofi-nccl/1.18.0-rocm (aws-ofi-nccl-1.18.0-rocm.eb)

Install with the EasyBuild-user module:

eb aws-ofi-nccl-1.18.0-rocm.eb -r

To access module help after installation and get reminded for which stacks and partitions the module is installed, use module spider aws-ofi-nccl/1.18.0-rocm.

EasyConfig:

easyblock = 'ConfigureMake'

name = 'aws-ofi-nccl'
version = '1.18.0'
versionsuffix = '-rocm'

homepage = 'https://github.com/aws/aws-ofi-nccl'

whatis = [
    'Description: AWS OFI NCCL is a plug-in which enables libfabric as a network provider while running NCCL/RCCL based applications.' 
]

description = """
Machine learning frameworks running on top of AMD GPUs use a library called 
RCCL which provides standard collective communication routines for an arbitrary 
number of GPUs installed across single or multiple nodes.

This module implements a plug-in which maps RCCLs connection-oriented transport 
APIs to libfabric's connection-less reliable interface. This allows RCCL 
applications to take benefit of libfabric's transport layer services like 
reliable message support and operating system bypass.
"""

# The plugin build needs access to MPI directory

toolchain = SYSTEM

sources = [{
    'filename': '%(name)s-%(version)s.tar.gz',
    'git_config': {
        'url': 'https://github.com/aws',
        'repo_name': '%(name)s',
        'commit': 'v%(version)s'
    }
}]

dependencies = [
    ('rocm', EXTERNAL_MODULE),
    ('libfabric/1.22.0', EXTERNAL_MODULE), # Cray branch
]

preconfigopts = ' ./autogen.sh && '
configopts = (
    ' CC=gcc-14 CXX=g++-14 '
    ' --with-libfabric=/opt/cray/libfabric/$(pkg-config --modversion libfabric) '
    ' --with-rocm=${ROCM_PATH} '
)

sanity_check_paths = {
    'files': ['lib/librccl-net.so', 'lib/librccl-net-ofi.so'],
    'dirs':  ['lib']
}

moduleclass = 'devel'

[aws-ofi-nccl] [package list]