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
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
|
# Copyright 2013-2021 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 *
from glob import glob
from llnl.util.filesystem import LibraryList
import os
import re
import platform
import llnl.util.tty as tty
# FIXME Remove hack for polymorphic versions
# This package uses a ugly hack to be able to dispatch, given the same
# version, to different binary packages based on the platform that is
# running spack. See #13827 for context.
# If you need to add a new version, please be aware that:
# - versions in the following dict are automatically added to the package
# - version tuple must be in the form (checksum, url)
# - checksum must be sha256
# - package key must be in the form '{os}-{arch}' where 'os' is in the
# format returned by platform.system() and 'arch' by platform.machine()
_versions = {
'11.3.1': {
'Linux-aarch64': ('39990d3da88b21289ac20850bc183f0b66275f32e1f562b551c05843bf506e4c', 'https://developer.download.nvidia.com/compute/cuda/11.3.1/local_installers/cuda_11.3.1_465.19.01_linux_sbsa.run'),
'Linux-x86_64': ('ad93ea98efced35855c58d3a0fc326377c60917cb3e8c017d3e6d88819bf2934', 'https://developer.download.nvidia.com/compute/cuda/11.3.1/local_installers/cuda_11.3.1_465.19.01_linux.run'),
'Linux-ppc64le': ('220f2c10a21500d62b03c6848c1659ebb3a8e10dc0915ab87b86b397058407c5', 'https://developer.download.nvidia.com/compute/cuda/11.3.1/local_installers/cuda_11.3.1_465.19.01_linux_ppc64le.run')},
'11.3.0': {
'Linux-aarch64': ('f7b284cf055fcf18be9a9aa216e3c7968d3e011446eb0c7200a3353c223ce718', 'https://developer.download.nvidia.com/compute/cuda/11.3.0/local_installers/cuda_11.3.0_465.19.01_linux_sbsa.run'),
'Linux-x86_64': ('262da7f77db177b153a6b70b5812210f3f72f00eb608ab9cf2a4582328c4bf5c', 'https://developer.download.nvidia.com/compute/cuda/11.3.0/local_installers/cuda_11.3.0_465.19.01_linux.run'),
'Linux-ppc64le': ('c0010107933b575a87e27b1293e5dc32b74201486f4ae2f4c8695ea727d22857', 'https://developer.download.nvidia.com/compute/cuda/11.3.0/local_installers/cuda_11.3.0_465.19.01_linux_ppc64le.run')},
'11.2.2': {
'Linux-aarch64': ('2f915ad631331eebdafaabd971723a60290ae8bb090d771075b9e6a0b28cbae6', 'https://developer.download.nvidia.com/compute/cuda/11.2.2/local_installers/cuda_11.2.2_460.32.03_linux_sbsa.run'),
'Linux-x86_64': ('0a2e477224af7f6003b49edfd2bfee07667a8148fe3627cfd2765f6ad72fa19d', 'https://developer.download.nvidia.com/compute/cuda/11.2.2/local_installers/cuda_11.2.2_460.32.03_linux.run'),
'Linux-ppc64le': ('2304ec235fe5d1f8bf75f00dc2c2d11473759dc23428dbbd5fb5040bc8c757e3', 'https://developer.download.nvidia.com/compute/cuda/11.2.2/local_installers/cuda_11.2.2_460.32.03_linux_ppc64le.run')},
'11.2.1': {
'Linux-aarch64': ('4b322fa6477d1a2cd2f2f526fa520c0f90bef2c264ef8435cb016bebb5456c5e', 'https://developer.download.nvidia.com/compute/cuda/11.2.1/local_installers/cuda_11.2.1_460.32.03_linux_sbsa.run'),
'Linux-x86_64': ('1da98cb897cc5f58a7445a4a66ca4f6926867706cb3af58a669cdcd8dc3d17c8', 'https://developer.download.nvidia.com/compute/cuda/11.2.1/local_installers/cuda_11.2.1_460.32.03_linux.run'),
'Linux-ppc64le': ('b3e8b6cd76872deb3acd050d32e197bc1c655e142b169070f0f9753680461a3f', 'https://developer.download.nvidia.com/compute/cuda/11.2.1/local_installers/cuda_11.2.1_460.32.03_linux_ppc64le.run')},
'11.2.0': {
'Linux-aarch64': ('c11dc274660e9b47b0f25ca66861a7406246a7191f1b04d0710515fcac0fa6cd', 'https://developer.download.nvidia.com/compute/cuda/11.2.0/local_installers/cuda_11.2.0_460.27.04_linux_sbsa.run'),
'Linux-x86_64': ('9c50283241ac325d3085289ed9b9c170531369de41165ce271352d4a898cbdce', 'https://developer.download.nvidia.com/compute/cuda/11.2.0/local_installers/cuda_11.2.0_460.27.04_linux.run'),
'Linux-ppc64le': ('adc3267df5dbfdaf51cb4c9b227ba6bfd979a39d9b4136bba0eba6b1dd2a2731', 'https://developer.download.nvidia.com/compute/cuda/11.2.0/local_installers/cuda_11.2.0_460.27.04_linux_ppc64le.run')},
'11.1.1': {
'Linux-aarch64': ('9ab1dbafba205c06bea8c88e38cdadb3038af19cb56e7b3ba734d3d7a84b8f02', 'https://developer.download.nvidia.com/compute/cuda/11.1.1/local_installers/cuda_11.1.1_455.32.00_linux_sbsa.run'),
'Linux-x86_64': ('3eae6727086024925ebbcef3e9a45ad379d8490768fd00f9c2d8b6fd9cd8dd8f', 'https://developer.download.nvidia.com/compute/cuda/11.1.1/local_installers/cuda_11.1.1_455.32.00_linux.run'),
'Linux-ppc64le': ('023e571fe26ee829c98138dfc305a92279854aac7d184d255fd58c06c6af3c17', 'https://developer.download.nvidia.com/compute/cuda/11.1.1/local_installers/cuda_11.1.1_455.32.00_linux_ppc64le.run')},
'11.1.0': {
'Linux-aarch64': ('878cbd36c5897468ef28f02da50b2f546af0434a8a89d1c724a4d2013d6aa993', 'https://developer.download.nvidia.com/compute/cuda/11.1.0/local_installers/cuda_11.1.0_455.23.05_linux_sbsa.run'),
'Linux-x86_64': ('858cbab091fde94556a249b9580fadff55a46eafbcb4d4a741d2dcd358ab94a5', 'https://developer.download.nvidia.com/compute/cuda/11.1.0/local_installers/cuda_11.1.0_455.23.05_linux.run'),
'Linux-ppc64le': ('a561e6f7f659bc4100e4713523b0b8aad6b36aa77fac847f6423e7780c750064', 'https://developer.download.nvidia.com/compute/cuda/11.1.0/local_installers/cuda_11.1.0_455.23.05_linux_ppc64le.run')},
'11.0.3': {
'Linux-aarch64': ('1e24f61f79c1043aa3d1d126ff6158daa03a62a51b5195a2ed5fbe75c3b718f3', 'https://developer.download.nvidia.com/compute/cuda/11.0.3/local_installers/cuda_11.0.3_450.51.06_linux_sbsa.run'),
'Linux-x86_64': ('b079c4e408adf88c3f1ffb8418a97dc4227c37935676b4bf4ca0beec6c328cc0', 'https://developer.download.nvidia.com/compute/cuda/11.0.3/local_installers/cuda_11.0.3_450.51.06_linux.run'),
'Linux-ppc64le': ('4775b21df004b1433bafff9b48a324075c008509f4c0fe28cd060d042d2e0794', 'https://developer.download.nvidia.com/compute/cuda/11.0.3/local_installers/cuda_11.0.3_450.51.06_linux_ppc64le.run')},
'11.0.2': {
'Linux-aarch64': ('23851e30f7c47a1baad92891abde0adbc783de5962c7480b9725198ceacda4a0', 'https://developer.download.nvidia.com/compute/cuda/11.0.2/local_installers/cuda_11.0.2_450.51.05_linux_sbsa.run'),
'Linux-x86_64': ('48247ada0e3f106051029ae8f70fbd0c238040f58b0880e55026374a959a69c1', 'https://developer.download.nvidia.com/compute/cuda/11.0.2/local_installers/cuda_11.0.2_450.51.05_linux.run'),
'Linux-ppc64le': ('db06d0f3fbf6f7aa1f106fc921ad1c86162210a26e8cb65b171c5240a3bf75da', 'https://developer.download.nvidia.com/compute/cuda/11.0.2/local_installers/cuda_11.0.2_450.51.05_linux_ppc64le.run')},
'10.2.89': {
'Linux-x86_64': ('560d07fdcf4a46717f2242948cd4f92c5f9b6fc7eae10dd996614da913d5ca11', 'https://developer.download.nvidia.com/compute/cuda/10.2/Prod/local_installers/cuda_10.2.89_440.33.01_linux.run'),
'Linux-ppc64le': ('5227774fcb8b10bd2d8714f0a716a75d7a2df240a9f2a49beb76710b1c0fc619', 'https://developer.download.nvidia.com/compute/cuda/10.2/Prod/local_installers/cuda_10.2.89_440.33.01_linux_ppc64le.run')},
'10.1.243': {
'Linux-x86_64': ('e7c22dc21278eb1b82f34a60ad7640b41ad3943d929bebda3008b72536855d31', 'https://developer.download.nvidia.com/compute/cuda/10.1/Prod/local_installers/cuda_10.1.243_418.87.00_linux.run'),
'Linux-ppc64le': ('b198002eef010bab9e745ae98e47567c955d00cf34cc8f8d2f0a6feb810523bf', 'https://developer.download.nvidia.com/compute/cuda/10.1/Prod/local_installers/cuda_10.1.243_418.87.00_linux_ppc64le.run')},
'10.0.130': {
'Linux-x86_64': ('92351f0e4346694d0fcb4ea1539856c9eb82060c25654463bfd8574ec35ee39a', 'https://developer.nvidia.com/compute/cuda/10.0/Prod/local_installers/cuda_10.0.130_410.48_linux')},
'9.2.88': {
'Linux-x86_64': ('8d02cc2a82f35b456d447df463148ac4cc823891be8820948109ad6186f2667c', 'https://developer.nvidia.com/compute/cuda/9.2/Prod/local_installers/cuda_9.2.88_396.26_linux')},
'9.1.85': {
'Linux-x86_64': ('8496c72b16fee61889f9281449b5d633d0b358b46579175c275d85c9205fe953', 'https://developer.nvidia.com/compute/cuda/9.1/Prod/local_installers/cuda_9.1.85_387.26_linux')},
'9.0.176': {
'Linux-x86_64': ('96863423feaa50b5c1c5e1b9ec537ef7ba77576a3986652351ae43e66bcd080c', 'https://developer.nvidia.com/compute/cuda/9.0/Prod/local_installers/cuda_9.0.176_384.81_linux-run')},
'8.0.61': {
'Linux-x86_64': ('9ceca9c2397f841024e03410bfd6eabfd72b384256fbed1c1e4834b5b0ce9dc4', 'https://developer.nvidia.com/compute/cuda/8.0/Prod2/local_installers/cuda_8.0.61_375.26_linux-run')},
'8.0.44': {
'Linux-x86_64': ('64dc4ab867261a0d690735c46d7cc9fc60d989da0d69dc04d1714e409cacbdf0', 'https://developer.nvidia.com/compute/cuda/8.0/prod/local_installers/cuda_8.0.44_linux-run')},
'7.5.18': {
'Linux-x86_64': ('08411d536741075131a1858a68615b8b73c51988e616e83b835e4632eea75eec', 'http://developer.download.nvidia.com/compute/cuda/7.5/Prod/local_installers/cuda_7.5.18_linux.run')},
'6.5.14': {
'Linux-x86_64': ('f3e527f34f317314fe8fcd8c85f10560729069298c0f73105ba89225db69da48', 'http://developer.download.nvidia.com/compute/cuda/6_5/rel/installers/cuda_6.5.14_linux_64.run')},
}
class Cuda(Package):
"""CUDA is a parallel computing platform and programming model invented
by NVIDIA. It enables dramatic increases in computing performance by
harnessing the power of the graphics processing unit (GPU).
Note: This package does not currently install the drivers necessary
to run CUDA. These will need to be installed manually. See:
https://docs.nvidia.com/cuda/ for details."""
homepage = "https://developer.nvidia.com/cuda-zone"
maintainers = ['ax3l', 'Rombur']
executables = ['^nvcc$']
for ver, packages in _versions.items():
key = "{0}-{1}".format(platform.system(), platform.machine())
pkg = packages.get(key)
if pkg:
version(ver, sha256=pkg[0], url=pkg[1], expand=False)
# macOS Mojave drops NVIDIA graphics card support -- official NVIDIA
# drivers do not exist for Mojave. See
# https://devtalk.nvidia.com/default/topic/1043070/announcements/faq-about-macos-10-14-mojave-nvidia-drivers/
# Note that a CUDA Toolkit installer does exist for macOS Mojave at
# https://developer.nvidia.com/compute/cuda/10.1/Prod1/local_installers/cuda_10.1.168_mac.dmg,
# but support for Mojave is dropped in later versions, and none of the
# macOS NVIDIA drivers at
# https://www.nvidia.com/en-us/drivers/cuda/mac-driver-archive/ mention
# Mojave support -- only macOS High Sierra 10.13 is supported.
conflicts('arch=darwin-mojave-x86_64')
variant('dev', default=False, description='Enable development dependencies, i.e to use cuda-gdb')
depends_on('libxml2', when='@10.1.243:')
# cuda-gdb needs libncurses.so.5
# see https://docs.nvidia.com/cuda/cuda-gdb/index.html#common-issues-oss
depends_on('ncurses abi=5', type='run', when='+dev')
provides('opencl@:1.2', when='@7:')
provides('opencl@:1.1', when='@:6')
@classmethod
def determine_version(cls, exe):
output = Executable(exe)('--version', output=str, error=str)
match = re.search(r'Cuda compilation tools, release .*?, V(\S+)',
output)
return match.group(1) if match else None
def setup_build_environment(self, env):
if self.spec.satisfies('@:8.0.61'):
# Perl 5.26 removed current directory from module search path,
# CUDA 9 has a fix for this, but CUDA 8 and lower don't.
env.append_path('PERL5LIB', self.stage.source_path)
if self.spec.satisfies('@10.1.243:'):
libxml2_home = self.spec['libxml2'].prefix
env.set('LIBXML2HOME', libxml2_home)
env.append_path('LD_LIBRARY_PATH', libxml2_home.lib)
def setup_dependent_build_environment(self, env, dependent_spec):
env.set('CUDAHOSTCXX', dependent_spec.package.compiler.cxx)
def setup_run_environment(self, env):
env.set('CUDA_HOME', self.prefix)
def install(self, spec, prefix):
if os.path.exists('/tmp/cuda-installer.log'):
try:
os.remove('/tmp/cuda-installer.log')
except OSError:
if spec.satisfies('@10.1:'):
tty.die("The cuda installer will segfault due to the "
"presence of /tmp/cuda-installer.log "
"please remove the file and try again ")
runfile = glob(join_path(self.stage.source_path, 'cuda*_linux*'))[0]
# Note: NVIDIA does not officially support many newer versions of
# compilers. For example, on CentOS 6, you must use GCC 4.4.7 or
# older. See:
# http://docs.nvidia.com/cuda/cuda-installation-guide-linux/#system-requirements
# https://gist.github.com/ax3l/9489132
# for details.
# CUDA 10.1 on ppc64le fails to copy some files, the workaround is adapted from
# https://forums.developer.nvidia.com/t/cuda-10-1-243-10-1-update-2-ppc64le-run-file-installation-issue/82433
# See also #21170
if spec.satisfies('@10.1.243') and platform.machine() == 'ppc64le':
includedir = "targets/ppc64le-linux/include"
os.makedirs(os.path.join(prefix, includedir))
os.makedirs(os.path.join(prefix, "src"))
os.symlink(includedir, os.path.join(prefix, "include"))
install_shell = which('sh')
if self.spec.satisfies('@:8.0.61'):
# Perl 5.26 removed current directory from module search path.
# We are addressing this by exporting `PERL5LIB` earlier, but for some
# reason, it is not enough. One more file needs to be extracted before
# running the actual installer. This solution is one of the commonly
# found on the Internet, when people try to install CUDA <= 8 manually.
# For example: https://askubuntu.com/a/1087842
arguments = [runfile, '--tar', 'mxvf', './InstallUtils.pm']
install_shell(*arguments)
# CUDA 10.1+ has different cmdline options for the installer
arguments = [
runfile, # the install script
'--silent', # disable interactive prompts
'--override', # override compiler version checks
'--toolkit', # install CUDA Toolkit
]
if spec.satisfies('@10.1:'):
arguments.append('--installpath=%s' % prefix) # Where to install
else:
arguments.append('--verbose') # Verbose log file
arguments.append('--toolkitpath=%s' % prefix) # Where to install
install_shell(*arguments)
try:
os.remove('/tmp/cuda-installer.log')
except OSError:
pass
@property
def libs(self):
libs = find_libraries('libcudart', root=self.prefix, shared=True,
recursive=True)
filtered_libs = []
# CUDA 10.0 provides Compatability libraries for running newer versions
# of CUDA with older drivers. These do not work with newer drivers.
for lib in libs:
parts = lib.split(os.sep)
if 'compat' not in parts and 'stubs' not in parts:
filtered_libs.append(lib)
return LibraryList(filtered_libs)
|
