• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1#!/bin/bash
2# Copyright 2022 Huawei Technologies Co., Ltd
3#
4# Licensed under the Apache License, Version 2.0 (the "License");
5# you may not use this file except in compliance with the License.
6# You may obtain a copy of the License at
7#
8# http://www.apache.org/licenses/LICENSE-2.0
9#
10# Unless required by applicable law or agreed to in writing, software
11# distributed under the License is distributed on an "AS IS" BASIS,
12# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13# See the License for the specific language governing permissions and
14# limitations under the License.
15# ============================================================================
16
17# Prepare and Install mindspore gpu by pip on Ubuntu 18.04.
18#
19# This file will:
20#   - change deb source to huaweicloud mirror
21#   - install mindspore dependencies via apt like gcc, libgmp
22#   - install python3 & pip3 via apt and set it to default
23#   - install CUDA by run file and cudnn via apt.
24#   - install mindspore-cpu within new installed python by pip
25#   - compile and install Open MPI if OPENMPI is set to on.
26#
27# Augments:
28#   - PYTHON_VERSION: python version to install. [3.7(default), 3.8, 3.9]
29#   - MINDSPORE_VERSION: mindspore version to install, >=1.6.0, required
30#   - CUDA_VERSION: CUDA version to install. [10.1, 11.1 11.6(default)]
31#   - OPENMPI: whether to install optional package Open MPI for distributed training. [on, off(default)]
32#
33# Usage:
34#   Run script like `MINDSPORE_VERSION=1.7.0 bash -i ./ubuntu-gpu-pip.sh`.
35#   To set augments, run it as `MINDSPORE_VERSION=1.6.0 CUDA_VERSION=10.1 OPENMPI=on bash -i ./ubuntu-gpu-pip.sh`.
36
37set -e
38
39PYTHON_VERSION=${PYTHON_VERSION:-3.7}
40MINDSPORE_VERSION=${MINDSPORE_VERSION:EMPTY}
41CUDA_VERSION=${CUDA_VERSION:-11.6}
42OPENMPI=${OPENMPI:-off}
43release_info=$(lsb_release -a | grep Release)
44UBUNTU_VERSION=${release_info//[!0-9]/}
45
46[[ "$UBUNTU_VERSION" == "2004" &&  "$CUDA_VERSION" == "10.1" ]] && echo "CUDA 10.1 is not supported on Ubuntu 20.04" && exit 1
47
48version_less() {
49    test "$(echo "$@" | tr ' ' '\n' | sort -rV | head -n 1)" != "$1";
50}
51
52if [ $MINDSPORE_VERSION == "EMPTY" ] || version_less "${MINDSPORE_VERSION}" "1.6.0"; then
53    echo "MINDSPORE_VERSION should be >=1.6.0, please check available versions at https://www.mindspore.cn/versions."
54    exit 1
55fi
56
57available_py_version=(3.7 3.8 3.9)
58if [[ " ${available_py_version[*]} " != *" $PYTHON_VERSION "* ]]; then
59    echo "PYTHON_VERSION is '$PYTHON_VERSION', but available versions are [${available_py_version[*]}]."
60    exit 1
61fi
62
63if [[ "$PYTHON_VERSION" == "3.8" && ${MINDSPORE_VERSION:0:3} == "1.6" ]]; then
64    echo "PYTHON_VERSION==3.8 is not compatible with MINDSPORE_VERSION==1.6.x, please use PYTHON_VERSION==3.7 or 3.9 for MINDSPORE_VERSION==1.6.x."
65    exit 1
66fi
67
68available_cuda_version=(10.1 11.1 11.6)
69if [[ " ${available_cuda_version[*]} " != *" $CUDA_VERSION "* ]]; then
70    echo "CUDA_VERSION is '$CUDA_VERSION', but available versions are [${available_cuda_version[*]}]."
71    exit 1
72fi
73declare -A minimum_driver_version_map=()
74minimum_driver_version_map["10.1"]="418.39"
75minimum_driver_version_map["11.1"]="450.80.02"
76minimum_driver_version_map["11.6"]="510.39.01"
77driver_version=$(nvidia-smi --query-gpu=driver_version --format=csv,noheader --id=0)
78if [[ $driver_version < ${minimum_driver_version_map[$CUDA_VERSION]} ]]; then
79    echo "CUDA $CUDA_VERSION minimum required driver version is ${minimum_driver_version_map[$CUDA_VERSION]}, \
80        but current nvidia driver version is $driver_version, please upgrade your driver manually."
81    exit 1
82fi
83cuda_name="cuda-$CUDA_VERSION"
84
85declare -A version_map=()
86version_map["3.7"]="${MINDSPORE_VERSION/-/}-cp37-cp37m"
87version_map["3.8"]="${MINDSPORE_VERSION/-/}-cp38-cp38"
88version_map["3.9"]="${MINDSPORE_VERSION/-/}-cp39-cp39"
89
90# add value to environment variable if value is not in it
91add_env() {
92    local name=$1
93    if [[ ":${!name}:" != *":$2:"* ]]; then
94        echo -e "export $1=$2:\$$1" >> ~/.bashrc
95    fi
96}
97
98# use huaweicloud mirror in China
99sudo sed -i "s@http://.*archive.ubuntu.com@http://repo.huaweicloud.com@g" /etc/apt/sources.list
100sudo sed -i "s@http://.*security.ubuntu.com@http://repo.huaweicloud.com@g" /etc/apt/sources.list
101sudo apt-get update
102
103sudo apt-get install curl make gcc-7 libgmp-dev -y
104
105# python
106sudo add-apt-repository -y ppa:deadsnakes/ppa
107sudo apt-get install python$PYTHON_VERSION python$PYTHON_VERSION-distutils python3-pip -y
108sudo update-alternatives --install /usr/bin/python python /usr/bin/python$PYTHON_VERSION 100
109# pip
110python -m pip install -U pip -i https://pypi.tuna.tsinghua.edu.cn/simple
111echo -e "alias pip='python -m pip'" >> ~/.bashrc
112python -m pip config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple
113
114# install cuda/cudnn
115echo "installing CUDA and cuDNN"
116cd /tmp
117declare -A cuda_url_map=()
118cuda_url_map["10.1"]=https://developer.download.nvidia.cn/compute/cuda/10.1/Prod/local_installers/cuda_10.1.243_418.87.00_linux.run
119cuda_url_map["11.1"]=https://developer.download.nvidia.cn/compute/cuda/11.1.1/local_installers/cuda_11.1.1_455.32.00_linux.run
120cuda_url_map["11.6"]=https://developer.download.nvidia.cn/compute/cuda/11.6.0/local_installers/cuda_11.6.0_510.39.01_linux.run
121cuda_url=${cuda_url_map[$CUDA_VERSION]}
122wget $cuda_url
123sudo sh ${cuda_url##*/} --silent --toolkit
124cd -
125sudo apt-key adv --fetch-keys https://developer.download.nvidia.cn/compute/cuda/repos/ubuntu${UBUNTU_VERSION}/x86_64/7fa2af80.pub
126sudo add-apt-repository "deb https://developer.download.nvidia.cn/compute/cuda/repos/ubuntu${UBUNTU_VERSION}/x86_64/ /"
127sudo add-apt-repository "deb https://developer.download.nvidia.cn/compute/machine-learning/repos/ubuntu${UBUNTU_VERSION}/x86_64/ /"
128sudo apt-get update
129declare -A cudnn_name_map=()
130cudnn_name_map["10.1"]="libcudnn7=7.6.5.32-1+cuda10.1 libcudnn7-dev=7.6.5.32-1+cuda10.1"
131cudnn_name_map["11.1"]="libcudnn8=8.0.5.39-1+cuda11.1 libcudnn8-dev=8.0.5.39-1+cuda11.1"
132cudnn_name_map["11.6"]="libcudnn8=8.5.0.96-1+cuda11.6 libcudnn8-dev=8.5.0.96-1+cuda11.6"
133sudo apt-get install --no-install-recommends ${cudnn_name_map[$CUDA_VERSION]} -y
134
135# add cuda to path
136set +e && source ~/.bashrc
137set -e
138add_env PATH /usr/local/cuda/bin
139add_env LD_LIBRARY_PATH /usr/local/cuda/lib64
140add_env LD_LIBRARY_PATH /usr/lib/x86_64-linux-gnu
141set +e && source ~/.bashrc
142set -e
143
144# optional openmpi for distributed training
145if [[ X"$OPENMPI" == "Xon" ]]; then
146    echo "installing openmpi"
147    cd /tmp
148    curl -O https://download.open-mpi.org/release/open-mpi/v4.1/openmpi-4.1.4.tar.gz
149    tar xzf openmpi-4.1.4.tar.gz
150    cd openmpi-4.1.4
151    ./configure --prefix=/usr/local/openmpi-4.1.4
152    make
153    sudo make install
154    add_env PATH /usr/local/openmpi-4.1.4/bin
155    add_env LD_LIBRARY_PATH /usr/local/openmpi-4.1.4/lib
156fi
157
158arch=`uname -m`
159python -m pip install https://ms-release.obs.cn-north-4.myhuaweicloud.com/${MINDSPORE_VERSION}/MindSpore/unified/${arch}/${cuda_name}/mindspore-${version_map["$PYTHON_VERSION"]}-linux_${arch}.whl --trusted-host ms-release.obs.cn-north-4.myhuaweicloud.com -i https://pypi.tuna.tsinghua.edu.cn/simple
160
161# check mindspore installation
162python -c "import mindspore;mindspore.run_check()"
163
164# check if it can be run with GPU
165cd /tmp
166cat > example.py <<END
167import numpy as np
168from mindspore import Tensor
169import mindspore.ops as ops
170import mindspore.context as context
171
172context.set_context(device_target="GPU")
173x = Tensor(np.ones([1,3,3,4]).astype(np.float32))
174y = Tensor(np.ones([1,3,3,4]).astype(np.float32))
175print(ops.add(x, y))
176END
177python example.py
178cd -
179