nvidia-apex-22.12.ebuild

This commit is contained in:
Anton Bolshakov 2022-12-01 15:16:38 +08:00
parent 692b486d96
commit 57542e4a92
No known key found for this signature in database
GPG key ID: 32BDCED870788F04
3 changed files with 50 additions and 78 deletions

View file

@ -1 +1 @@
DIST nvidia-apex-22.08-gh.tar.gz 746501 BLAKE2B 7d882e551c14b7b3a6b463f7873d22603b55e957ffd792735ab05ba158706c61c492325f87fb5d3370cf96108a7a4ec9546efc93c78a3e96f16470e43b93cee8 SHA512 11291444f5038f6b1702e7bd2c65daa29746ada9add18d1af226ed7d3facd21c4fe4ba2e262936f6a95e3a84f41184edf2a1ff964394fc6eb7b23e76157dcdf5
DIST nvidia-apex-22.12-gh.tar.gz 790427 BLAKE2B 2b53e1c2ded28d6f3de8812bd6e8ed1db35d10b3293ef53ea7df63eebe5474f2b8e8af6c5700e9f19581715cb73de893b2736944ef621f2eff37959650641784 SHA512 418c76f655d0cb1cfebc8b007b914a2b8e3d9d54efd6496911702577913f7887bc744228e3a465c13c77bcccbf7971f9f9a5a2bd217e395abb92b1fdfbc18ff2

View file

@ -1,77 +0,0 @@
# Copyright 1999-2022 Gentoo Authors
# Distributed under the terms of the GNU General Public License v2
EAPI=8
DISTUTILS_IN_SOURCE_BUILD=1
PYTHON_COMPAT=( python3_{10..11} )
inherit distutils-r1
#git branch
HASH_COMMIT="${PV}-dev"
DESCRIPTION="NVIDIA-maintained utilities to streamline mixed precision and distributed training in Pytorch"
HOMEPAGE="https://github.com/NVIDIA/apex"
SRC_URI="https://github.com/NVIDIA/apex/archive/${HASH_COMMIT}.tar.gz -> ${P}-gh.tar.gz"
LICENSE=""
SLOT="0"
# unable to build in sanbox for some reason
#KEYWORDS="amd64 ~arm64 x86"
#FIXME: can't use global "cuda"
IUSE="cuda_ext"
RDEPEND=">=dev-python/tqdm-4.28.1[${PYTHON_USEDEP}]
>=dev-python/numpy-1.15.3[${PYTHON_USEDEP}]
>=dev-python/pyyaml-5.1[${PYTHON_USEDEP}]"
DEPEND="${RDEPEND}"
REQUIRED_USE="${PYTHON_REQUIRED_USE}"
RESTRICT="test"
S="${WORKDIR}/apex-${HASH_COMMIT}"
#python_prepare() {
# if use !cuda_ext; then
# einfo "CUDA disabled"
#sed -i -e "s|fused_layer_norm_cuda = importlib|\#fused_layer_norm_cuda = importlib|" apex/normalization/fused_layer_norm.py || die
# eapply ${FILESDIR}/disable_cuda.patch
# fi
# eapply_user
#}
#python_prepare_all() {
# export TORCH_CUDA_ARCH_LIST="compute capability"
# python_setup
# esetup.py
# distutils-r1_python_prepare_all
#}
#If you wish to cross-compile for a single specific architecture,
#export TORCH_CUDA_ARCH_LIST="compute capability" before running setup.py.
python_configure_all() {
# export MAX_JOBS=1
if use cuda_ext; then
# export TORCH_CUDA_ARCH_LIST="compute capability"
addpredict "/dev/nvidiactl"
DISTUTILS_ARGS=( --cpp_ext --cuda_ext )
fi
}
#python_compile_all() {
# export TORCH_CUDA_ARCH_LIST="compute capability"
# esetup.py
#}
#python_compile() {
# export TORCH_CUDA_ARCH_LIST="compute capability"
# breaks with parallel build
# need to avoid dropping .so plugins into
# build-lib, which breaks tests
# esetup.py build_ext --inplace
# --cpp_ext --cuda_ext build_ext --inplace -j1
# TORCH_CUDA_ARCH_LIST="compute capability" distutils-r1_python_compile -j1
# distutils-r1_python_compile -j1
#}

View file

@ -0,0 +1,49 @@
# Copyright 1999-2022 Gentoo Authors
# Distributed under the terms of the GNU General Public License v2
EAPI=8
PYTHON_COMPAT=( python3_{10..11} )
inherit distutils-r1
#git branch
HASH_COMMIT="${PV}-dev"
DESCRIPTION="NVIDIA-maintained utilities to streamline mixed precision and distributed training in Pytorch"
HOMEPAGE="https://github.com/NVIDIA/apex"
SRC_URI="https://github.com/NVIDIA/apex/archive/${HASH_COMMIT}.tar.gz -> ${P}-gh.tar.gz"
LICENSE=""
SLOT="0"
KEYWORDS="amd64 ~arm64 x86"
#FIXME: can't use global "cuda"
IUSE="cuda_ext"
RDEPEND=">=dev-python/cxxfilt-0.2.0[${PYTHON_USEDEP}]
>=dev-python/tqdm-4.28.1[${PYTHON_USEDEP}]
>=dev-python/numpy-1.15.3[${PYTHON_USEDEP}]
>=dev-python/pyyaml-5.1[${PYTHON_USEDEP}]
>=dev-python/packaging-14.0[${PYTHON_USEDEP}]"
DEPEND="${RDEPEND}"
REQUIRED_USE="${PYTHON_REQUIRED_USE}"
RESTRICT="test"
S="${WORKDIR}/apex-${HASH_COMMIT}"
#If you wish to cross-compile for a single specific architecture,
#export TORCH_CUDA_ARCH_LIST="compute capability" before running setup.py.
python_configure_all() {
if use cuda_ext; then
# export MAX_JOBS=1
#export TORCH_CUDA_ARCH_LIST="compute capability"
export TORCH_CUDA_ARCH_LIST="7.5"
DISTUTILS_ARGS=( --cpp_ext --cuda_ext )
fi
}
python_compile() {
distutils-r1_python_compile -j1
}