
We'd like to use a consistent checksum scheme everywhere so that we can: a) incorporate archive checksums into our specs and have a consistent hashing algorithm across all specs. b) index mirrors with a consistent type of checksum, and not one that is dependent on how spack packages are written. - [x] convert existing md5, sha224, sha512, sha1 checksums to sha256
38 lines
1.6 KiB
Python
38 lines
1.6 KiB
Python
# Copyright 2013-2019 Lawrence Livermore National Security, LLC and other
|
|
# Spack Project Developers. See the top-level COPYRIGHT file for details.
|
|
#
|
|
# SPDX-License-Identifier: (Apache-2.0 OR MIT)
|
|
|
|
from spack import *
|
|
|
|
|
|
class Nccl(MakefilePackage):
|
|
"""Optimized primitives for collective multi-GPU communication."""
|
|
|
|
homepage = "https://github.com/NVIDIA/nccl"
|
|
url = "https://github.com/NVIDIA/nccl/archive/v2.4.8-1.tar.gz"
|
|
|
|
version('2.4.8-1', sha256='e2260da448ebbebe437f74768a346d28c74eabdb92e372a3dc6652a626318924')
|
|
version('2.4.6-1', sha256='ea4421061a7b9c454f2e088f68bfdbbcefab80ce81cafc70ee6c7742b1439591')
|
|
version('2.4.2-1', sha256='e3dd04b22eb541394bd818e5f78ac23a09cc549690d5d55d6fccc1a36155385a')
|
|
version('2.3.7-1', sha256='e6eff80d9d2db13c61f8452e1400ca2f098d2dfe42857cb23413ce081c5b9e9b')
|
|
version('2.3.5-5', sha256='bac9950b4d3980c25baa8e3e4541d2dfb4d21edf32ad3b89022d04920357142f')
|
|
version('1.3.4-1', sha256='11e4eb44555bb28b9cbad973dacb4640b82710c9769e719afc2013b63ffaf884')
|
|
version('1.3.0-1', sha256='53f36151061907bdcafad1c26c1d9370a0a8400f561a83704a5138213ba51003')
|
|
|
|
depends_on('cuda')
|
|
|
|
# https://github.com/NVIDIA/nccl/issues/244
|
|
patch('so_reuseport.patch', when='@2.3.7-1:2.4.8-1')
|
|
|
|
@property
|
|
def build_targets(self):
|
|
return ['CUDA_HOME={0}'.format(self.spec['cuda'].prefix)]
|
|
|
|
@property
|
|
def install_targets(self):
|
|
if self.version >= Version('2.3.5-5'):
|
|
return ['PREFIX={0}'.format(self.prefix), 'src.install']
|
|
else:
|
|
return ['PREFIX={0}'.format(self.prefix), 'install']
|