zhangyiss/spack
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

py-horovod: added versions up to v0.20.3 (#18977)

Browse Source
This commit is contained in:
Adam J. Stewart 2020-10-05 09:22:33 -05:00 committed by GitHub
parent cb7a1ac2bf
commit 0d64020007
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 52 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

65
var/spack/repos/builtin/packages/py-horovod/package.py
View File

@ -4,16 +4,20 @@
# SPDX-License-Identifier: (Apache-2.0 OR MIT) # SPDX-License-Identifier: (Apache-2.0 OR MIT)
class PyHorovod(PythonPackage): class PyHorovod(PythonPackage, CudaPackage):
"""Horovod is a distributed deep learning training framework for """Horovod is a distributed deep learning training framework for
TensorFlow, Keras, PyTorch, and Apache MXNet.""" TensorFlow, Keras, PyTorch, and Apache MXNet."""
homepage = "https://github.com/horovod" homepage = "https://github.com/horovod"
git = "https://github.com/horovod/horovod.git" git = "https://github.com/horovod/horovod.git"
maintainers = ['adamjstewart'] maintainers = ['adamjstewart', 'aweits', 'tgaddair']
version('master', branch='master', submodules=True) version('master', branch='master', submodules=True)
version('0.20.3', tag='v0.20.3', submodules=True)
version('0.20.2', tag='v0.20.2', submodules=True)
version('0.20.1', tag='v0.20.1', submodules=True)
version('0.20.0', tag='v0.20.0', submodules=True)
version('0.19.5', tag='v0.19.5', submodules=True) version('0.19.5', tag='v0.19.5', submodules=True)
version('0.19.4', tag='v0.19.4', submodules=True) version('0.19.4', tag='v0.19.4', submodules=True)
version('0.19.3', tag='v0.19.3', submodules=True) version('0.19.3', tag='v0.19.3', submodules=True)
@ -32,7 +36,7 @@ class PyHorovod(PythonPackage):
# https://github.com/horovod/horovod/blob/master/docs/install.rst # https://github.com/horovod/horovod/blob/master/docs/install.rst
variant('frameworks', default='pytorch', variant('frameworks', default='pytorch',
description='Deep learning frameworks to build support for', description='Deep learning frameworks to build support for',
values=('tensorflow', 'pytorch', 'mxnet', 'keras', 'spark'), values=('tensorflow', 'pytorch', 'mxnet', 'keras', 'spark', 'ray'),
multi=True) multi=True)
variant('controllers', default='mpi', variant('controllers', default='mpi',
description='Controllers to coordinate work between processes', description='Controllers to coordinate work between processes',
@ -40,17 +44,23 @@ class PyHorovod(PythonPackage):
variant('tensor_ops', default='nccl', variant('tensor_ops', default='nccl',
description='Framework to use for GPU/CPU operations', description='Framework to use for GPU/CPU operations',
values=('nccl', 'mpi', 'gloo', 'ccl'), multi=False) values=('nccl', 'mpi', 'gloo', 'ccl'), multi=False)
variant('cuda', default=True, description='Build with CUDA')
variant('rocm', default=False, description='Build with ROCm')
# Required dependencies # Required dependencies
depends_on('python@3.6:', type=('build', 'run'), when='@0.20:')
depends_on('py-setuptools', type='build') depends_on('py-setuptools', type='build')
depends_on('py-cloudpickle', type=('build', 'run')) depends_on('py-cloudpickle', type=('build', 'run'))
depends_on('py-psutil', type=('build', 'run')) depends_on('py-psutil', type=('build', 'run'))
depends_on('py-pyyaml', type=('build', 'run')) depends_on('py-pyyaml', type=('build', 'run'))
depends_on('py-six', type=('build', 'run')) depends_on('py-six', type=('build', 'run'), when='@:0.19')
depends_on('py-dataclasses', type=('build', 'run'), when='@0.20: ^python@:3.6')
# Framework dependencies # Framework dependencies
depends_on('py-tensorflow@1.1.0:', type=('build', 'link', 'run'), when='frameworks=tensorflow') depends_on('py-tensorflow@1.1.0:', type=('build', 'link', 'run'), when='frameworks=tensorflow')
depends_on('py-tensorflow@1.15:', type=('build', 'link', 'run'), when='frameworks=tensorflow @0.20:')
depends_on('py-torch@0.4.0:', type=('build', 'link', 'run'), when='frameworks=pytorch') depends_on('py-torch@0.4.0:', type=('build', 'link', 'run'), when='frameworks=pytorch')
depends_on('py-torch@1.2:', type=('build', 'link', 'run'), when='frameworks=pytorch @0.20:')
depends_on('py-torchvision', type=('build', 'run'), when='frameworks=pytorch @:0.19.1') depends_on('py-torchvision', type=('build', 'run'), when='frameworks=pytorch @:0.19.1')
depends_on('py-cffi@1.4.0:', type=('build', 'run'), when='frameworks=pytorch') depends_on('py-cffi@1.4.0:', type=('build', 'run'), when='frameworks=pytorch')
depends_on('mxnet@1.4.1:+python', type=('build', 'link', 'run'), when='frameworks=mxnet') depends_on('mxnet@1.4.1:+python', type=('build', 'link', 'run'), when='frameworks=mxnet')
@ -61,11 +71,17 @@ class PyHorovod(PythonPackage):
depends_on('py-petastorm@0.9.0:', type=('build', 'run'), when='frameworks=spark @0.19.2:') depends_on('py-petastorm@0.9.0:', type=('build', 'run'), when='frameworks=spark @0.19.2:')
depends_on('py-pyarrow@0.15.0:', type=('build', 'run'), when='frameworks=spark') depends_on('py-pyarrow@0.15.0:', type=('build', 'run'), when='frameworks=spark')
depends_on('py-pyspark@2.3.2:', type=('build', 'run'), when='frameworks=spark') depends_on('py-pyspark@2.3.2:', type=('build', 'run'), when='frameworks=spark')
depends_on('py-ray', type=('build', 'run'), when='frameworks=ray')
# Build dependencies
depends_on('cmake@2.8.12:', type='build', when='@0.20:')
depends_on('pkgconfig', type='build')
# Controller dependencies # Controller dependencies
depends_on('mpi', when='controllers=mpi') depends_on('mpi', when='controllers=mpi')
# There does not appear to be a way to use an external Gloo installation # There does not appear to be a way to use an external Gloo installation
depends_on('cmake', type='build', when='controllers=gloo') depends_on('cmake', type='build', when='controllers=gloo')
depends_on('libuv@1.26:', when='controllers=gloo platform=darwin')
# Tensor Operations dependencies # Tensor Operations dependencies
depends_on('nccl@2:', when='tensor_ops=nccl') depends_on('nccl@2:', when='tensor_ops=nccl')
@ -77,13 +93,28 @@ class PyHorovod(PythonPackage):
depends_on('py-mock', type='test') depends_on('py-mock', type='test')
depends_on('py-pytest', type='test') depends_on('py-pytest', type='test')
depends_on('py-pytest-forked', type='test') depends_on('py-pytest-forked', type='test')
depends_on('py-parameterized', type='test', when='@0.20:')
conflicts('controllers=gloo', when='platform=darwin', msg='Gloo cannot be compiled on MacOS') conflicts('cuda_arch=none', when='+cuda',
msg='Must specify CUDA compute capabilities of your GPU, see '
'https://developer.nvidia.com/cuda-gpus')
conflicts('tensor_ops=nccl', when='~cuda~rocm', msg='NCCL requires either CUDA or ROCm support')
conflicts('framework=ray', when='@:0.19', msg='Ray integration was added in 0.20.X')
conflicts('controllers=gloo', when='@:0.20.0 platform=darwin', msg='Gloo cannot be compiled on MacOS')
# https://github.com/horovod/horovod/pull/1835 # https://github.com/horovod/horovod/pull/1835
patch('fma.patch', when='@0.19.0:0.19.1') patch('fma.patch', when='@0.19.0:0.19.1')
def setup_build_environment(self, env): def setup_build_environment(self, env):
# https://github.com/horovod/horovod/blob/master/docs/install.rst#environment-variables
# Build system
env.set('PKG_CONFIG_EXECUTABLE',
self.spec['pkgconfig'].prefix.bin.join('pkg-config'))
if '^cmake' in self.spec:
env.set('HOROVOD_CMAKE', self.spec['cmake'].command.path)
env.set('MAKEFLAGS', '-j{0}'.format(make_jobs))
# Frameworks # Frameworks
if 'frameworks=tensorflow' in self.spec: if 'frameworks=tensorflow' in self.spec:
env.set('HOROVOD_WITH_TENSORFLOW', 1) env.set('HOROVOD_WITH_TENSORFLOW', 1)
@ -110,20 +141,28 @@ def setup_build_environment(self, env):
# Tensor Operations # Tensor Operations
if 'tensor_ops=nccl' in self.spec: if 'tensor_ops=nccl' in self.spec:
env.set('HOROVOD_GPU', 'CUDA') env.set('HOROVOD_GPU_ALLREDUCE', 'NCCL')
env.set('HOROVOD_GPU_ALLGATHER', 'NCCL')
env.set('HOROVOD_CUDA_HOME', self.spec['cuda'].prefix) env.set('HOROVOD_GPU_BROADCAST', 'NCCL')
env.set('HOROVOD_CUDA_INCLUDE',
self.spec['cuda'].headers.directories[0])
env.set('HOROVOD_CUDA_LIB', self.spec['cuda'].libs.directories[0])
env.set('HOROVOD_NCCL_HOME', self.spec['nccl'].prefix) env.set('HOROVOD_NCCL_HOME', self.spec['nccl'].prefix)
env.set('HOROVOD_NCCL_INCLUDE', env.set('HOROVOD_NCCL_INCLUDE',
self.spec['nccl'].headers.directories[0]) self.spec['nccl'].headers.directories[0])
env.set('HOROVOD_NCCL_LIB', self.spec['nccl'].libs.directories[0]) env.set('HOROVOD_NCCL_LIB', self.spec['nccl'].libs.directories[0])
env.set('HOROVOD_GPU_ALLREDUCE', 'NCCL') if '+cuda' in self.spec:
env.set('HOROVOD_GPU_BROADCAST', 'NCCL') env.set('HOROVOD_GPU', 'CUDA')
env.set('HOROVOD_CUDA_HOME', self.spec['cuda'].prefix)
cuda_cc_list = ','.join(self.spec.variants['cuda_arch'].value)
env.set('HOROVOD_BUILD_CUDA_CC_LIST', cuda_cc_list)
env.set('HOROVOD_CUDA_INCLUDE',
self.spec['cuda'].headers.directories[0])
env.set('HOROVOD_CUDA_LIB',
self.spec['cuda'].libs.directories[0])
elif '+rocm' in self.spec:
env.set('HOROVOD_GPU', 'ROCM')
# env.set('HOROVOD_ROCM_HOME', self.spec['rocm'].prefix)
else: else:
env.set('HOROVOD_CPU_OPERATIONS', env.set('HOROVOD_CPU_OPERATIONS',
self.spec.variants['tensor_ops'].value.upper()) self.spec.variants['tensor_ops'].value.upper())