记一次失败记录: MindSpore1.3.0 GPU 源码安装 —— Ubuntu18.04系统 (最终安装结果为失败)
官网地址:
https://www.mindspore.cn/install
系统:Ubuntu18.04
硬件:
i7-9700k CPU
2060super nvidia显卡
由于考虑到mindspore版本更新,依赖环境的变化,因此将mindspore依赖的环境安装到自建文件夹下,
给出下面操作中环境的安装文件夹:
/home/devilmaycry/anaconda3/mindspore_envs
这个路径可以按照个人情况进行设置的,只要下面均保持一致即可。
下文中修改 .bashrc 是指打开.bashrc :
vim ~/.bashrc
本文假设系统中已经为GPU显卡安装好驱动程序了,因此本文不介绍驱动的安装。
(请注意:下面的安装建议按照步骤来进行)
================================================================
1. 安装CUDA11.1.0 和 cuDNN 8.0.X版本:
cuda安装
创建cuda的安装目录:
mkdir cuda11.1
查看所创建路径:
/home/devilmaycry/anaconda3/mindspore_envs/cuda11.1
下载并安装:
wget https://developer.download.nvidia.com/compute/cuda/11.1.0/local_installers/cuda_11.1.0_455.23.05_linux.run
sh ./cuda_11.1.0_455.23.05_linux.run --toolkit --toolkitpath=/home/devilmaycry/anaconda3/mindspore_envs/cuda11.1 --silent
2. cuDnn安装:
下载地址:
解压文件:
tar -zxvf cudnn-11.1-linux-x64-v8.0.5.39.tgz
拷贝解压后的文件到cuda安装目录内:
cp cuda/include/* /home/devilmaycry/anaconda3/mindspore_envs/cuda11.1/include
cp cuda/lib64/* /home/devilmaycry/anaconda3/mindspore_envs/cuda11.1/lib64
配置环境变量:
修改 .bashrc 文件
# cuda&cudnn export LD_LIBRARY_PATH=/home/devilmaycry/anaconda3/mindspore_envs/cuda11.1/lib64:$LD_LIBRARY_PATH export PATH=/home/devilmaycry/anaconda3/mindspore_envs/cuda11.1/bin:$PATH
================================================================
GCC的安装:
下载gcc 7.3.0版本安装包,执行以下命令:
wget http://ftp.gnu.org/gnu/gcc/gcc-7.3.0/gcc-7.3.0.tar.gz
执行tar -xzf gcc-7.3.0.tar.gz
解压源码包。
执行cd gcc-7.3.0
,进入到源码包目录。
继续下面操作前清空系统内的环境变量:
export LIBRARY_PATH= export LD_LIBRARY_PATH= export C_INCLUDE_PATH= export CPLUS_INCLUDE_PATH=
运行以下命令,进行安装前的配置。
安装依赖环境:
./contrib/download_prerequisites
配置环境:
./configure --prefix=/home/devilmaycry/anaconda3/mindspore_envs/gcc_7.3.0 --enable-bootstrap -enable-threads=posix --enable-checking=release --enable-languages=c,c++ --disable-multilib
编译安装:
make -j8 && make install
配置系统环境: 修改 .bashrc 文件,添加内容:
# gcc export PATH=/home/devilmaycry/anaconda3/mindspore_envs/gcc_7.3.0/bin:$PATH ###export LIBRARY_PATH=/home/devilmaycry/anaconda3/mindspore_envs/gcc_7.3.0/share:$LIBRARY_PATH export LD_LIBRARY_PATH=/home/devilmaycry/anaconda3/mindspore_envs/gcc_7.3.0/lib64:$LD_LIBRARY_PATH export C_INCLUDE_PATH=/home/devilmaycry/anaconda3/mindspore_envs/gcc_7.3.0/include:$C_INCLUDE_PATH export CPLUS_INCLUDE_PATH=$C_INCLUDE_PATH:$CPLUS_INCLUDE_PATH
=========================================================
m4 下载安装:
下载:
wget http://ftp.gnu.org/gnu/m4/m4-1.4.16.tar.bz2
解压:
tar -jxvf m4-1.4.16.tar.bz2
修改m4_1.4.16下源文件中代码:(https://blog.csdn.net/weixin_34168880/article/details/91842744)
vi lib/stdio.in.h
查找字段:gets is a security hole
注释:
将_GL_WARN_ON_USE (gets, "gets is a security hole - use fgets instead"); 字段和他之前的注释 /* 一块注释掉,如下
/* It is very rare that the developer ever has full control of stdin, so any use of gets warrants an unconditional warning. Assume it is always declared, since it is required by C89. #undef gets _GL_WARN_ON_USE (gets, "gets is a security hole - use fgets instead"); */
再添加如下内容:
#if defined(__GLIBC__) && !defined(__UCLIBC__) && !__GLIBC_PREREQ(2, 16) _GL_WARN_ON_USE (gets, "gets is a security hole - use fgets instead"); #endif
配置:
./configure --prefix=/home/devilmaycry/anaconda3/mindspore_envs/m4_1.4.16/
编译安装
make && make install
配置系统环境变量:
修改 .bashrc 文件
export PATH=/home/devilmaycry/anaconda3/mindspore_envs/m4_1.4.16/bin:$PATH
说明: 其中,/home/devilmaycry/anaconda3/mindspore_envs/m4_1.4.16 路径为我们刚才编译源码配置安装路径的地址。
==============================================================
需要注意:安装好m4后需要执行命令 soure ~/.bashrc ,然后才可以继续安装gmp。
安装gmp 6.1.2
下载gmp 6.1.2
源码包:
wget https://gmplib.org/download/gmp/gmp-6.1.2.tar.xz
解压到当前文件夹:
tar -xvf gmp-6.1.2.tar.xz
配置:
./configure --prefix=/home/devilmaycry/anaconda3/mindspore_envs/gmp_6.1.2/ --enable-cxx
编译安装:
make && make install
配置系统环境变量: 修改 .bashrc 文件
# gmp export LIBRARY_PATH=/home/devilmaycry/anaconda3/mindspore_envs/gmp_6.1.2/lib:$LIBRARY_PATH export LD_LIBRARY_PATH=/home/devilmaycry/anaconda3/mindspore_envs/gmp_6.1.2/lib:$LD_LIBRARY_PATH export C_INCLUDE_PATH=/home/devilmaycry/anaconda3/mindspore_envs/gmp_6.1.2/include:$C_INCLUDE_PATH export CPLUS_INCLUDE_PATH=$C_INCLUDE_PATH:$CPLUS_INCLUDE_PATH
测试 gmp 是否安装并配置成功:(声明:测试部分内容源于:https://blog.csdn.net/just_h/article/details/82667787)
代码:
# test.cpp 文件
#include <gmpxx.h> #include <iostream> #include <stdio.h> using namespace std; int main() { mpz_t a,b,c; mpz_init(a); mpz_init(b); mpz_init(c); gmp_scanf("%Zd%Zd",a,b); mpz_add(c,a,b); gmp_printf("c= %Zd\n",c); return 0; }
编译:
g++ test.cpp -o test -lgmp
运行:
openssl 的安装:
下载地址:
https://www.openssl.org/source/openssl-1.1.1k.tar.gz
解压:
tar -zxvf openssl-1.1.1k
配置:
./config --prefix=/home/devilmaycry/anaconda3/mindspore_envs/openssl_1.1.1
编译并安装:
make -j8&& make install
配置系统环境:修改 .bashrc 文件, 添加内容:
# openssl export OPENSSL_ROOT_DIR=/home/devilmaycry/anaconda3/mindspore_envs/openssl_1.1.1 export OPENSSL_INCLUDE_DIR=/home/devilmaycry/anaconda3/mindspore_envs/openssl_1.1.1/include export OPENSSL_CRYPTO_LIBRARY=/home/devilmaycry/anaconda3/mindspore_envs/openssl_1.1.1/lib
重新载入 .bashrc 文件:
source ~/.bashrc
cmake 安装:
下载文件:
wget https://github.com/Kitware/CMake/releases/download/v3.21.0/cmake-3.21.0.tar.gz
解压:
tar -zxvf cmake-3.21.0.tar.gz
配置:
./configure --prefix=/home/devilmaycry/anaconda3/mindspore_envs/cmake_3.21.0
编译并安装:
make -j8&& make install
配置系统环境:
修改 ~/.bashrc 文件,添加内容:
# make export PATH=/home/devilmaycry/anaconda3/mindspore_envs/cmake_3.21.0/bin:$PATH
重新载入 .bashrc 文件:
source ~/.bashrc
patch 的安装:
下载文件:
wget https://ftp.gnu.org/gnu/patch/patch-2.7.6.tar.gz
解压:
tar -zxvf patch-2.7.6.tar.gz
配置:
./configure --prefix=/home/devilmaycry/anaconda3/mindspore_envs/patch_2.7.6
编译并安装:
make -j8&& make install
配置系统环境:
修改 ~/.bashrc 文件,添加内容:
# patch
export PATH=/home/devilmaycry/anaconda3/mindspore_envs/patch_2.7.6/bin:$PATH
重新载入 .bashrc 文件:
source ~/.bashrc
安装Autoconf
下载文件:
wget https://ftp.gnu.org/gnu/autoconf/autoconf-2.71.tar.gz
解压:
tar -zxvf autoconf-2.71.tar.gz
配置:
./configure --prefix=/home/devilmaycry/anaconda3/mindspore_envs/autoconf_2.71
编译并安装:
make -j8&& make install
配置系统环境:
修改 ~/.bashrc 文件,添加内容:
# autoconf export PATH=/home/devilmaycry/anaconda3/mindspore_envs/autoconf_2.71/bin:$PATH
重新载入 .bashrc 文件:
source ~/.bashrc
安装 libtool
下载地址:
wget https://ftpmirror.gnu.org/libtool/libtool-2.4.6.tar.gz
解压:
tar -zxvf libtool-2.4.6.tar.gz
配置:
./configure --prefix=/home/devilmaycry/anaconda3/mindspore_envs/libtool_2.4.6
编译并安装:
make -j8&& make install
配置系统环境:
修改 ~/.bashrc 文件,添加内容:
# libtool export PATH=/home/devilmaycry/anaconda3/mindspore_envs/libtool_2.4.6/bin:$PATH
重新载入 .bashrc 文件:
source ~/.bashrc
安装automake
下载:
wget https://ftp.gnu.org/gnu/automake/automake-1.16.3.tar.gz
解压:
tar -zxvf automake-1.16.3.tar.gz
配置:
./configure --prefix=/home/devilmaycry/anaconda3/mindspore_envs/automake_1.16.3
编译并安装:
make -j8&& make install
配置系统环境:
修改 ~/.bashrc 文件,添加内容:
# automake export PATH=/home/devilmaycry/anaconda3/mindspore_envs/automake_1.16.3/bin:$PATH
重新载入 .bashrc 文件:
source ~/.bashrc
安装flex
下载文件:
wget https://github.com/westes/flex/files/981163/flex-2.6.4.tar.gz
解压:
tar -zxvf flex-2.6.4.tar.gz
配置:
./configure --prefix=/home/devilmaycry/anaconda3/mindspore_envs/flex_2.6.4
编译并安装:
make -j8&& make install
修改系统环境,修改 .bashrc文件,添加内容:
# flex export PATH=/home/devilmaycry/anaconda3/mindspore_envs/flex_2.6.4/bin:$PATH export LIBRARY_PATH=/home/devilmaycry/anaconda3/mindspore_envs/flex_2.6.4/lib:$LIBRARY_PATH export LD_LIBRARY_PATH=/home/devilmaycry/anaconda3/mindspore_envs/flex_2.6.4/lib:$LD_LIBRARY_PATH export C_INCLUDE_PATH=/home/devilmaycry/anaconda3/mindspore_envs/flex_2.6.4/include:$C_INCLUDE_PATH export CPLUS_INCLUDE_PATH=$C_INCLUDE_PATH:$CPLUS_INCLUDE_PATH
重新载入 .bashrc 文件:
source ~/.bashrc
安装Python环境和wheel
这里我们使用anaconda下的Python环境,anaconda的安装这里不进行讲解:
创建Python环境,命名mindspore_1.3.0:
conda create -n mindspore_1.3.0 python=3.7.5
安装 wheel :
激活Python环境:
activate mindspore_1.3.0
安装:
pip install wheel
安装NUMA
下载:
wget https://github.com/numactl/numactl/releases/download/v2.0.14/numactl-2.0.14.tar.gz
解压:
tar -zxvf numactl-2.0.14.tar.gz
配置:
./configure --prefix=/home/devilmaycry/anaconda3/mindspore_envs/numactl_2.0.14
编译并安装:
make -j8&& make install
修改系统环境,修改 .bashrc文件,添加内容:
# numa export PATH=/home/devilmaycry/anaconda3/mindspore_envs/numactl_2.0.14/bin:$PATH export LIBRARY_PATH=/home/devilmaycry/anaconda3/mindspore_envs/numactl_2.0.14/lib:$LIBRARY_PATH export LD_LIBRARY_PATH=/home/devilmaycry/anaconda3/mindspore_envs/numactl_2.0.14/lib:$LD_LIBRARY_PATH export C_INCLUDE_PATH=/home/devilmaycry/anaconda3/mindspore_envs/numactl_2.0.14/include:$C_INCLUDE_PATH export CPLUS_INCLUDE_PATH=$C_INCLUDE_PATH:$CPLUS_INCLUDE_PATH
重新载入 .bashrc 文件:
source ~/.bashrc
=========================================================================
必要环境已经全部安装好,正式编译:
git clone https://gitee.com/mindspore/mindspore.git -b r1.3
bash build.sh -e gpu
失败。
报错信息:
---------------- MindSpore: build start ----------------
mkdir: 已创建目录 '/tmp/mindspore/build'
mkdir: 已创建目录 '/tmp/mindspore/build//package'
mkdir: 已创建目录 '/tmp/mindspore/build//package/mindspore'
mkdir: 已创建目录 '/tmp/mindspore/build//package/mindspore/lib'
子模组 'graphengine'(https://gitee.com/mindspore/graphengine.git)未对路径 'graphengine' 注册
正克隆到 '/tmp/mindspore/graphengine'...
子模组路径 'graphengine':检出 'acc2472c4119d70b579d7ac2f7a6bbf2c344af29'
子模组 'metadef'(https://gitee.com/ascend/metadef.git)未对路径 'metadef' 注册
正克隆到 '/tmp/mindspore/graphengine/metadef'...
子模组路径 'metadef':检出 '0a9ebe1c7f7b27554659f39e387110ac30d4a1e6'
子模组 'akg'(https://gitee.com/mindspore/akg.git)未对路径 'akg' 注册
正克隆到 '/tmp/mindspore/akg'...
子模组路径 'akg':检出 '796260887e9c87964aad87ab8154211060870fec'
start build mindspore project.
mkdir: 已创建目录 '/tmp/mindspore/build//mindspore'
-DDEBUG_MODE=off -DBUILD_PATH=/tmp/mindspore/build/ -DENABLE_INFER=ON
-DENABLE_DUMP_PROTO=ON -DENABLE_DUMP_IR=on -DENABLE_PYTHON=on
-DENABLE_MPI=ON -DENABLE_GPU=ON -DUSE_CUDA=ON -DCUDA_PATH=
-DMS_REQUIRE_CUDA_VERSION=10.1 -DENABLE_CPU=ON -DX86_64_SIMD=off
-DARM_SIMD=off -DENABLE_MINDDATA=ON -DUSE_GLOG=ON -DENABLE_AKG=ON
-DENABLE_DEBUGGER=ON
-- The C compiler identification is GNU 7.3.0
-- The CXX compiler identification is GNU 7.3.0
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: /home/devilmaycry/anaconda3/mindspore_envs/gcc_7.3.0/bin/gcc - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: /home/devilmaycry/anaconda3/mindspore_envs/gcc_7.3.0/bin/g++ - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Found Python3:
/home/devilmaycry/anaconda3/envs/mindspore_1.3.0/bin/python3.7 (found
version "3.7.5") found components: Interpreter Development
Development.Module Development.Embed
Python3 found, version: 3.7.5
Python3 library path:
Python3 interpreter: /home/devilmaycry/anaconda3/envs/mindspore_1.3.0/bin/python3.7
-- Found Patch: /home/devilmaycry/anaconda3/mindspore_envs/patch_2.7.6/bin/patch
PATCH_EXECUTABLE=/home/devilmaycry/anaconda3/mindspore_envs/patch_2.7.6/bin/patch
-- Looking for pthread.h
-- Looking for pthread.h - found
-- Performing Test CMAKE_HAVE_LIBC_PTHREAD
-- Performing Test CMAKE_HAVE_LIBC_PTHREAD - Failed
-- Looking for pthread_create in pthreads
-- Looking for pthread_create in pthreads - not found
-- Looking for pthread_create in pthread
-- Looking for pthread_create in pthread - found
-- Found Threads: TRUE
CMake Error at cmake/check_requirements.cmake:66 (message):
Required package gmp not found, please install gmp and try building
MindSpore again.
Call Stack (most recent call first):
CMakeLists.txt:13 (include)
-- Configuring incomplete, errors occurred!
See also "/tmp/mindspore/build/mindspore/CMakeFiles/CMakeOutput.log".
See also "/tmp/mindspore/build/mindspore/CMakeFiles/CMakeError.log".
参考:
https://gitee.com/mindspore/mindspore/issues/I284ZL
========================================================================
另附其他环境的安装方式:
mpfr 的安装:
下载地址:
wget https://www.mpfr.org/mpfr-current/mpfr-4.1.0.tar.gz
解压文件:
tar -zxvf mpfr-4.1.0.tar.gz
配置:
./configure --prefix=/home/devilmaycry/anaconda3/mindspore_envs/mpfr_4.1.0/
编译安装:
make && make install
配置文件: 添加内容到 .bashrc 文件中
# mpfr
export LIBRARY_PATH=/home/devilmaycry/anaconda3/mindspore_envs/mpfr_4.1.0/lib:$LIBRARY_PATH
export LD_LIBRARY_PATH=/home/devilmaycry/anaconda3/mindspore_envs/mpfr_4.1.0/lib:$LD_LIBRARY_PATH
export C_INCLUDE_PATH=/home/devilmaycry/anaconda3/mindspore_envs/mpfr_4.1.0/include:$C_INCLUDE_PATH
export CPLUS_INCLUDE_PATH=$C_INCLUDE_PATH:$CPLUS_INCLUDE_PATH
测试安装及配置是否成功:
代码: #test.cpp
#include <stdio.h>
#include <mpfr.h>
int main (void)
{
printf ("MPFR library: %-12s\nMPFR header: %s (based on %d.%d.%d)\n",
mpfr_get_version (), MPFR_VERSION_STRING, MPFR_VERSION_MAJOR,
MPFR_VERSION_MINOR, MPFR_VERSION_PATCHLEVEL);
return 0;
}
编译及运行:
gcc test.cpp -o test -lmpfr
mpc 的安装:
下载:
wget https://ftp.gnu.org/gnu/mpc/mpc-1.2.1.tar.gz
解压:
tar -zxvf mpc-1.2.1.tar.gz
配置:
./configure --prefix=/home/devilmaycry/anaconda3/mindspore_envs/mpc_1.2.1/
编译安装:
make && make install
修改 .bashrc 文件,添加内容:
# mpc
export LIBRARY_PATH=/home/devilmaycry/anaconda3/mindspore_envs/mpc_1.2.1/share:$LIBRARY_PATH
export LD_LIBRARY_PATH=/home/devilmaycry/anaconda3/mindspore_envs/mpc_1.2.1/lib:$LD_LIBRARY_PATH
export C_INCLUDE_PATH=/home/devilmaycry/anaconda3/mindspore_envs/mpc_1.2.1/include:$C_INCLUDE_PATH
export CPLUS_INCLUDE_PATH=$C_INCLUDE_PATH:$CPLUS_INCLUDE_PATH
==================================================================
posted on 2021-07-21 22:58 Angry_Panda 阅读(637) 评论(2) 编辑 收藏 举报