1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
|
# Copyright 2013-2020 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 Dihydrogen(CMakePackage, CudaPackage):
"""DiHydrogen is the second version of the Hydrogen fork of the
well-known distributed linear algebra library,
Elemental. DiHydrogen aims to be a basic distributed
multilinear algebra interface with a particular emphasis on the
needs of the distributed machine learning effort, LBANN."""
homepage = "https://github.com/LLNL/DiHydrogen.git"
url = "https://github.com/LLNL/DiHydrogen.git"
git = "https://github.com/LLNL/DiHydrogen.git"
maintainers = ['bvanessen']
version('master', branch='master')
variant('al', default=True,
description='Builds with Aluminum communication library')
variant('developer', default=False,
description='Enable extra warnings and force tests to be enabled.')
variant('half', default=False,
description='Enable FP16 support on the CPU.')
variant('legacy', default=False,
description='Enable the legacy DistConv code branch.')
variant('nvshmem', default=False,
description='Builds with support for NVSHMEM')
variant('openmp', default=False,
description='Enable CPU acceleration with OpenMP threads.')
variant('rocm', default=False,
description='Enable ROCm/HIP language features.')
variant('shared', default=True,
description='Enables the build of shared libraries')
variant('docs', default=False,
description='Builds with support for building documentation')
# Variants related to BLAS
variant('openmp_blas', default=False,
description='Use OpenMP for threading in the BLAS library')
variant('int64_blas', default=False,
description='Use 64bit integers for BLAS.')
variant('blas', default='openblas', values=('openblas', 'mkl', 'accelerate', 'essl'),
description='Enable the use of OpenBlas/MKL/Accelerate/ESSL')
# Override the default set of CUDA architectures with the relevant
# subset from lib/spack/spack/build_systems/cuda.py
cuda_arch_values = [
'60', '61', '62',
'70', '72', '75',
'80'
]
variant('cuda_arch',
description='CUDA architecture',
values=spack.variant.auto_or_any_combination_of(*cuda_arch_values))
depends_on('mpi')
depends_on('catch2', type='test')
depends_on('aluminum', when='+al ~cuda')
depends_on('aluminum +cuda +nccl +ht +cuda_rma', when='+al +cuda')
depends_on('cuda', when=('+cuda' or '+legacy'))
depends_on('cudnn', when=('+cuda' or '+legacy'))
depends_on('cub', when=('+cuda' or '+legacy'))
# Note that #1712 forces us to enumerate the different blas variants
depends_on('openblas', when='blas=openblas ~openmp_blas ~int64_blas')
depends_on('openblas +ilp64', when='blas=openblas ~openmp_blas +int64_blas')
depends_on('openblas threads=openmp', when='blas=openblas +openmp_blas ~int64_blas')
depends_on('openblas threads=openmp +lip64', when='blas=openblas +openmp_blas +int64_blas')
depends_on('intel-mkl', when="blas=mkl ~openmp_blas ~int64_blas")
depends_on('intel-mkl +ilp64', when="blas=mkl ~openmp_blas +int64_blas")
depends_on('intel-mkl threads=openmp', when='blas=mkl +openmp_blas ~int64_blas')
depends_on('intel-mkl@2017.1 +openmp +ilp64', when='blas=mkl +openmp_blas +int64_blas')
depends_on('veclibfort', when='blas=accelerate')
conflicts('blas=accelerate +openmp_blas')
depends_on('essl -cuda', when='blas=essl -openmp_blas ~int64_blas')
depends_on('essl -cuda +ilp64', when='blas=essl -openmp_blas +int64_blas')
depends_on('essl threads=openmp', when='blas=essl +openmp_blas ~int64_blas')
depends_on('essl threads=openmp +ilp64', when='blas=essl +openmp_blas +int64_blas')
depends_on('netlib-lapack +external-blas', when='blas=essl')
# Legacy builds require cuda
conflicts('~cuda', when='+legacy')
depends_on('half', when='+half')
generator = 'Ninja'
depends_on('ninja', type='build')
depends_on('cmake@3.16.0:', type='build')
depends_on('py-breathe', type='build', when='+docs')
depends_on('doxygen', type='build', when='+docs')
illegal_cuda_arch_values = [
'10', '11', '12', '13',
'20', '21',
'30', '32', '35', '37',
'50', '52', '53',
]
for value in illegal_cuda_arch_values:
conflicts('cuda_arch=' + value)
@property
def libs(self):
shared = True if '+shared' in self.spec else False
return find_libraries(
'libH2Core', root=self.prefix, shared=shared, recursive=True
)
def cmake_args(self):
spec = self.spec
args = [
'-DCMAKE_INSTALL_MESSAGE:STRING=LAZY',
'-DBUILD_SHARED_LIBS:BOOL=%s' % ('+shared' in spec),
'-DH2_ENABLE_CUDA=%s' % ('+cuda' in spec),
'-DH2_ENABLE_DISTCONV_LEGACY=%s' % ('+legacy' in spec),
'-DH2_ENABLE_OPENMP=%s' % ('+openmp' in spec),
'-DH2_ENABLE_FP16=%s' % ('+half' in spec),
'-DH2_ENABLE_HIP_ROCM=%s' % ('+rocm' in spec),
'-DH2_DEVELOPER_BUILD=%s' % ('+developer' in spec),
]
if '+cuda' in spec:
cuda_arch = spec.variants['cuda_arch'].value
if len(cuda_arch) == 1 and cuda_arch[0] == 'auto':
args.append('-DCMAKE_CUDA_FLAGS=-arch=sm_60')
else:
cuda_arch = [x for x in spec.variants['cuda_arch'].value
if x != 'auto']
if cuda_arch:
args.append('-DCMAKE_CUDA_FLAGS={0}'.format(
' '.join(self.cuda_flags(cuda_arch))
))
return args
|
