91 lines
4.1 KiB
Python
91 lines
4.1 KiB
Python
# 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 PyTorchvision(PythonPackage):
|
|
"""The torchvision package consists of popular datasets, model
|
|
architectures, and common image transformations for computer vision."""
|
|
|
|
homepage = "https://github.com/pytorch/vision"
|
|
url = "https://github.com/pytorch/vision/archive/v0.7.0.tar.gz"
|
|
|
|
maintainers = ['adamjstewart']
|
|
import_modules = [
|
|
'torchvision', 'torchvision.datasets', 'torchvision.models',
|
|
'torchvision.transforms', 'torchvision.ops',
|
|
'torchvision.models.segmentation',
|
|
'torchvision.models.detection'
|
|
]
|
|
|
|
version('0.7.0', sha256='fa0a6f44a50451115d1499b3f2aa597e0092a07afce1068750260fa7dd2c85cb')
|
|
version('0.6.1', sha256='8173680a976c833640ecbd0d7e6f0a11047bf8833433e2147180efc905e48656')
|
|
version('0.6.0', sha256='02de11b3abe6882de4032ce86dab9c7794cbc84369b44d04e667486580f0f1f7')
|
|
version('0.5.0', sha256='eb9afc93df3d174d975ee0914057a9522f5272310b4d56c150b955c287a4d74d')
|
|
version('0.4.2', sha256='1184a27eab85c9e784bacc6f9d6fec99e168ab4eda6047ef9f709e7fdb22d8f9')
|
|
version('0.4.1', sha256='053689351272b3bd2ac3e6ba51efd284de0e4ca4a301f54674b949f1e62b7176')
|
|
version('0.4.0', sha256='c270d74e568bad4559fed4544f6dd1e22e2eb1c60b088e04a5bd5787c4150589')
|
|
version('0.3.0', sha256='c205f0618c268c6ed2f8abb869ef6eb83e5339c1336c243ad321a2f2a85195f0')
|
|
|
|
# See README.rst
|
|
depends_on('python@3.6:', when='@0.7:', type=('build', 'run'))
|
|
depends_on('python@3.5:', when='@0.6.0:0.6.999', type=('build', 'run'))
|
|
depends_on('python@2.7:2.8,3.5:3.8', when='@0.5.0', type=('build', 'run'))
|
|
depends_on('python@2.7:2.8,3.5:3.7', when='@:0.4', type=('build', 'run'))
|
|
|
|
depends_on('py-setuptools', type='build')
|
|
depends_on('ninja', type='build')
|
|
depends_on('py-numpy', type=('build', 'run'))
|
|
depends_on('py-six', when='@:0.5', type=('build', 'run'))
|
|
|
|
# See README.rst
|
|
depends_on('py-torch@1.6.0', when='@0.7.0', type=('build', 'link', 'run'))
|
|
depends_on('py-torch@1.5.1', when='@0.6.1', type=('build', 'link', 'run'))
|
|
depends_on('py-torch@1.5.0', when='@0.6.0', type=('build', 'link', 'run'))
|
|
depends_on('py-torch@1.4.0', when='@0.5.0', type=('build', 'link', 'run'))
|
|
depends_on('py-torch@1.3.1', when='@0.4.2', type=('build', 'link', 'run'))
|
|
depends_on('py-torch@1.3.0', when='@0.4.1', type=('build', 'link', 'run'))
|
|
depends_on('py-torch@1.2.0', when='@0.4.0', type=('build', 'link', 'run'))
|
|
depends_on('py-torch@1.1.0', when='@0.3.0', type=('build', 'link', 'run'))
|
|
depends_on('py-torch@:1.0.1', when='@0.2.2', type=('build', 'link', 'run'))
|
|
|
|
# TODO: Torchvision supports the following backends:
|
|
# * pillow
|
|
# * pillow-simd
|
|
# * accimage
|
|
# * libpng
|
|
# * libjpeg
|
|
|
|
# https://github.com/pytorch/vision/issues/1712
|
|
depends_on('py-pillow@4.1.1:6', when='@:0.4', type=('build', 'run'))
|
|
depends_on('py-pillow@4.1.1:', when='@0.5:', type=('build', 'run'))
|
|
|
|
# Many of the datasets require additional dependencies to use.
|
|
# These can be installed after the fact.
|
|
depends_on('py-scipy', type='test')
|
|
|
|
depends_on('ffmpeg@3.1:', when='@0.4.2:')
|
|
|
|
def setup_build_environment(self, env):
|
|
include = []
|
|
library = []
|
|
for dep in self.spec.dependencies(deptype='link'):
|
|
query = self.spec[dep.name]
|
|
include.extend(query.headers.directories)
|
|
library.extend(query.libs.directories)
|
|
|
|
# README says to use TORCHVISION_INCLUDE and TORCHVISION_LIBRARY,
|
|
# but these do not work. Build uses a mix of Spack's compiler wrapper
|
|
# and the actual compiler, so this is needed to get parts of the build
|
|
# working. See https://github.com/pytorch/vision/issues/2591
|
|
env.set('CPATH', ':'.join(include))
|
|
env.set('LIBRARY_PATH', ':'.join(library))
|
|
|
|
if '+cuda' in self.spec['py-torch']:
|
|
env.set('FORCE_CUDA', 1)
|
|
else:
|
|
env.set('FORCE_CUDA', 0)
|