mirror of
https://github.com/pentoo/pentoo-overlay
synced 2026-05-05 11:00:48 +02:00
nvidia-apex, WIP
This commit is contained in:
Anton Bolshakov
parent
5ba1f511e0
commit
829c6b24b0
8 changed files with 117 additions and 56 deletions
1
dev-python/cxxfilt/Manifest
Normal file
1
dev-python/cxxfilt/Manifest
Normal file
|
|
@ -0,0 +1 @@
|
|||
DIST cxxfilt-0.3.0.tar.gz 4806 BLAKE2B 369f10c547343dc9e4bf87edd4513531109b1792d5196da57fd0a5db9f0830bf5ada1f20dd7799a3a114400a3f03cbfc795c021d2ae763557e848abca4f7094e SHA512 87cf42c8ae7a2eeb74712b2d952d011ca1e2ac90e3d9c7d254eeef1dd88ceba1f25aff377852d384f73c38cd0183898f3ba9ccde0d9f09d03025c0262290704f
|
||||
24
dev-python/cxxfilt/cxxfilt-0.3.0.ebuild
Normal file
24
dev-python/cxxfilt/cxxfilt-0.3.0.ebuild
Normal file
|
|
@ -0,0 +1,24 @@
|
|||
# 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
|
||||
|
||||
DESCRIPTION="Python interface to c++filt / abi::__cxa_demangle"
|
||||
HOMEPAGE="https://github.com/afq984/python-cxxfilt"
|
||||
SRC_URI="mirror://pypi/${P:0:1}/${PN}/${P}.tar.gz"
|
||||
|
||||
LICENSE="BSD-2"
|
||||
SLOT="0"
|
||||
KEYWORDS="amd64 ~arm64 x86"
|
||||
IUSE=""
|
||||
RESTRICT="test"
|
||||
|
||||
RDEPEND=""
|
||||
DEPEND="${RDEPEND}"
|
||||
REQUIRED_USE="${PYTHON_REQUIRED_USE}"
|
||||
|
||||
#distutils_enable_tests pytest
|
||||
1
dev-python/nvidia-apex-bin/Manifest
Normal file
1
dev-python/nvidia-apex-bin/Manifest
Normal file
|
|
@ -0,0 +1 @@
|
|||
DIST nvidia-apex-bin-22.8.zip 3247310 BLAKE2B ef1d8ff2933d5b9866b137645769e9f1ccda2091e1dfd73b5a86e6556ef5af927e652e36181b425e2d0406f8dcd51ce0a65be86eb4d2deb278cfaf83e8d63d43 SHA512 0b92db6858f075b007fc46ce794cb015d4ad018e787850d2c25e0669f4f39ef67d1efbf5baa14d139ef747fae94ffbbf1673c62652a038fcde983178701e740d
|
||||
37
dev-python/nvidia-apex-bin/nvidia-apex-bin-22.8.ebuild
Normal file
37
dev-python/nvidia-apex-bin/nvidia-apex-bin-22.8.ebuild
Normal file
|
|
@ -0,0 +1,37 @@
|
|||
# Copyright 1999-2022 Gentoo Authors
|
||||
# Distributed under the terms of the GNU General Public License v2
|
||||
|
||||
EAPI=8
|
||||
|
||||
PYTHON_COMPAT=( python3_10 )
|
||||
|
||||
#inherit distutils-r1
|
||||
inherit python-any-r1
|
||||
|
||||
#22.08 -> 22.8
|
||||
MY_P="apex-${PV}-py3.10-linux-x86_64.egg"
|
||||
|
||||
DESCRIPTION="NVIDIA-maintained utilities to streamline mixed precision and distributed training in Pytorch"
|
||||
HOMEPAGE="https://github.com/NVIDIA/apex"
|
||||
SRC_URI="https://dev.pentoo.ch/~blshkv/distfiles/nvidia-${MY_P} -> ${P}.zip"
|
||||
|
||||
LICENSE=""
|
||||
SLOT="0"
|
||||
#KEYWORDS="amd64 ~arm64 x86"
|
||||
IUSE=""
|
||||
|
||||
#[${PYTHON_USEDEP}]
|
||||
RDEPEND=">=dev-python/tqdm-4.28.1
|
||||
>=dev-python/numpy-1.15.3
|
||||
>=dev-python/pyyaml-5.1
|
||||
"
|
||||
DEPEND="${RDEPEND}"
|
||||
|
||||
RESTRICT="test"
|
||||
|
||||
S=${WORKDIR}
|
||||
|
||||
src_install() {
|
||||
insinto "$(python_get_sitedir)/${MY_P}"
|
||||
doins -r "./"
|
||||
}
|
||||
|
|
@ -1,2 +1 @@
|
|||
DIST nvidia-apex-22.04-gh.tar.gz 769456 BLAKE2B 9c9270e5b19afa244784f0825a2c093820aab5e64f5621ea0350da6b127d14b734fcbf16982ff889a82b624842d982c0f447146a9efddd165dd59916988ee955 SHA512 347a67640dd1903442dacc59cb7c15efe299cafc3d7afd36ae2f1fbe4b7b08d23c18a5351035f62f3b7f47023331d37485a99cc0049ef8c788f27a32146d9e64
|
||||
DIST nvidia-apex-22.08-gh.tar.gz 746501 BLAKE2B 7d882e551c14b7b3a6b463f7873d22603b55e957ffd792735ab05ba158706c61c492325f87fb5d3370cf96108a7a4ec9546efc93c78a3e96f16470e43b93cee8 SHA512 11291444f5038f6b1702e7bd2c65daa29746ada9add18d1af226ed7d3facd21c4fe4ba2e262936f6a95e3a84f41184edf2a1ff964394fc6eb7b23e76157dcdf5
|
||||
|
|
|
|||
11
dev-python/nvidia-apex/files/disable_cuda.patch
Normal file
11
dev-python/nvidia-apex/files/disable_cuda.patch
Normal file
|
|
@ -0,0 +1,11 @@
|
|||
--- apex-22.08-dev/apex/normalization/fused_layer_norm.py.orig 2022-08-02 08:25:04.000000000 +0800
|
||||
+++ apex-22.08-dev/apex/normalization/fused_layer_norm.py 2022-11-21 13:12:25.579343523 +0800
|
||||
@@ -265,7 +265,7 @@
|
||||
super().__init__()
|
||||
|
||||
global fused_layer_norm_cuda
|
||||
- fused_layer_norm_cuda = importlib.import_module("fused_layer_norm_cuda")
|
||||
+ #fused_layer_norm_cuda = importlib.import_module("fused_layer_norm_cuda")
|
||||
|
||||
if isinstance(normalized_shape, numbers.Integral):
|
||||
normalized_shape = (normalized_shape,)
|
||||
|
|
@ -1,37 +0,0 @@
|
|||
# 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
|
||||
|
||||
#brach 22.04-dev
|
||||
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 -> ${P}-gh.tar.gz"
|
||||
|
||||
LICENSE=""
|
||||
SLOT="0"
|
||||
KEYWORDS="amd64 ~arm64 x86"
|
||||
IUSE=""
|
||||
|
||||
RDEPEND=">=dev-python/tqdm-4.28.1[${PYTHON_USEDEP}]
|
||||
>=dev-python/numpy-1.15.3[${PYTHON_USEDEP}]
|
||||
>=dev-python/pyyaml-5.1[${PYTHON_USEDEP}]"
|
||||
DEPEND="${RDEPEND}"
|
||||
REQUIRED_USE="${PYTHON_REQUIRED_USE}"
|
||||
|
||||
S="${WORKDIR}/apex-${HASH_COMMIT}"
|
||||
|
||||
#src_compile(){
|
||||
# pip install -v --disable-pip-version-check --no-cache-dir ./
|
||||
#pip install -v --disable-pip-version-check --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./
|
||||
#}
|
||||
|
||||
python_compile() {
|
||||
distutils-r1_python_compile --cpp_ext --cuda_ext
|
||||
}
|
||||
|
|
@ -3,11 +3,12 @@
|
|||
|
||||
EAPI=8
|
||||
|
||||
DISTUTILS_IN_SOURCE_BUILD=1
|
||||
PYTHON_COMPAT=( python3_{10..11} )
|
||||
|
||||
inherit distutils-r1
|
||||
|
||||
#brach 22.04-dev
|
||||
#git branch
|
||||
HASH_COMMIT="${PV}-dev"
|
||||
|
||||
DESCRIPTION="NVIDIA-maintained utilities to streamline mixed precision and distributed training in Pytorch"
|
||||
|
|
@ -17,7 +18,9 @@ SRC_URI="https://github.com/NVIDIA/apex/archive/${HASH_COMMIT}.tar.gz -> ${P}-gh
|
|||
LICENSE=""
|
||||
SLOT="0"
|
||||
KEYWORDS="amd64 ~arm64 x86"
|
||||
IUSE="cuda"
|
||||
|
||||
#FIXME: can't use global "cuda"
|
||||
IUSE="cuda_ext"
|
||||
|
||||
RDEPEND=">=dev-python/tqdm-4.28.1[${PYTHON_USEDEP}]
|
||||
>=dev-python/numpy-1.15.3[${PYTHON_USEDEP}]
|
||||
|
|
@ -25,27 +28,49 @@ RDEPEND=">=dev-python/tqdm-4.28.1[${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_compile() {
|
||||
if use cuda; then
|
||||
distutils-r1_python_compile --cpp_ext --cuda_ext
|
||||
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_install() {
|
||||
if use cuda; then
|
||||
# disable gpu check
|
||||
# required for cross-compile and sanbox
|
||||
export TORCH_CUDA_ARCH_LIST="compute capability"
|
||||
distutils-r1_python_install --cpp_ext --cuda_ext
|
||||
fi
|
||||
}
|
||||
#python_compile_all() {
|
||||
# export TORCH_CUDA_ARCH_LIST="compute capability"
|
||||
# esetup.py
|
||||
#}
|
||||
|
||||
|
||||
#FIXME:
|
||||
#https://github.com/NVIDIA/apex/issues/161
|
||||
# "No module named 'fused_layer_norm_cuda'"
|
||||
#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
|
||||
#}
|
||||
|
|
|
|||
Loading…
Reference in a new issue