Turbot-DL入门教程篇-深度学习框架Torch7安装
说明：
介绍如何在turbot-DL安装深度学习框架Torch7
步骤：
新建安装脚本：
mkdir -p ~/dl/Torch7 cd ~/dl/Torch7 sudo vim jetson_torch7_install.sh
内容如下：
#!/usr/bin/env bash
# this script is automatically run from CMakeLists.txt
BUILD_ROOT=$PWD
BUILD_OPENBLAS=$1
BUILD_PYTORCH=$2
BUILD_TORCH=$3
TORCH_PREFIX=$PWD/torch
echo "[Pre-build] dependency installer script running..."
echo "[Pre-build] BUILD_ROOT directory: $BUILD_ROOT"
echo "[Pre-build] BUILD_OPENBLAS $BUILD_OPENBLAS"
echo "[Pre-build] BUILD_PYTORCH $BUILD_PYTORCH"
echo "[Pre-build] BUILD_TORCH $BUILD_TORCH"
echo "[Pre-build] installing Torch/LUA into: $TORCH_PREFIX"
# (don't) break on errors
#set -e
#
# (prompt) install Gazebo7
#
while true; do
read -p "[Pre-build] Do you wish to install Gazebo robotics simulator (y/N)? " yn
case $yn in
[Yy]* ) sudo apt-get update; sudo apt-get install -y gazebo7 libgazebo7-dev; break;;
[Nn]* ) echo "[Pre-build] skipping Gazebo installation"; break;;
* ) echo "[Pre-build] Please answer yes or no.";;
esac
done
#
# build pyTorch?
#
if [ $BUILD_PYTORCH = "ON" ] || [ $BUILD_PYTORCH = "YES" ] || [ $BUILD_PYTORCH = "Y" ]; then
echo "[Pre-build] beginning pyTorch setup"
sudo apt-get install python-pip
# upgrade pip
pip install -U pip
pip --version
# pip 9.0.1 from /home/ubuntu/.local/lib/python2.7/site-packages (python 2.7)
# clone pyTorch repo
git clone http://github.com/pytorch/pytorch
cd pytorch
# install prereqs
sudo pip install -U setuptools
sudo pip install -r requirements.txt
# Develop Mode:
python setup.py build_deps
sudo python setup.py develop
git clone http://github.com/pytorch/examples
sudo pip install -r examples/reinforcement_learning/requirements.txt
cd ../
echo "[Pre-build] pyTorch setup complete"
fi
#
# build Torch?
#
if [ $BUILD_TORCH = "ON" ] || [ $BUILD_TORCH = "YES" ] || [ $BUILD_TORCH = "Y" ]; then
# Install dependencies for Torch:
echo "[Pre-build] installing Torch7 package dependencies"
sudo apt-get update
sudo apt-get install -y gfortran build-essential gcc g++ cmake curl libreadline-dev git-core liblua5.1-0-dev
# note: gfortran is for OpenBLAS LAPACK compilation
##sudo apt-get install -qqy libgd-dev
sudo apt-get update
echo "[Pre-build] Torch7's package dependencies have been installed"
# Install openBLAS
echo "[Pre-build] build OpenBLAS? $1"
if [ $1 = "ON" ] || [ $1 = "YES" ] || [ $1 = "Y" ]; then
echo "[Pre-build] building OpenBLAS...";
rm -rf OpenBLAS
git clone http://github.org/xianyi/OpenBLAS
cd OpenBLAS
mkdir build
make
make install PREFIX=$BUILD_ROOT/OpenBLAS/build
export CMAKE_LIBRARY_PATH=$BUILD_ROOT/OpenBLAS/build/include:$BUILD_ROOT/OpenBLAS/build/lib:$CMAKE_LIBRARY_PATH
cd $BUILD_ROOT
fi
# Install luajit-rocks into Torch dir
echo "[Pre-build] configuring luajit-rocks"
cd $BUILD_ROOT
rm -rf luajit-rocks
git clone https://github.com/torch/luajit-rocks.git
cd luajit-rocks
mkdir -p build
cd build
git checkout master; git pull
rm -f CMakeCache.txt
cmake .. -DWITH_LUAJIT21=yes -DCMAKE_INSTALL_PREFIX=$TORCH_PREFIX -DCMAKE_BUILD_TYPE=Release
make
make install
cd $BUILD_ROOT
path_to_nvcc=$(which nvcc)
if [ -x "$path_to_nvcc" ]
then
cutorch=ok
cunn=ok
echo "[Pre-build] detected NVCC / CUDA toolkit"
fi
sudo apt-get install gnuplot gnuplot-qt
# Install base packages:
echo "[Pre-build] installing luarocks packages"
cd $BUILD_ROOT
git clone http://github.com/torch/rocks
$TORCH_PREFIX/bin/luarocks install $BUILD_ROOT/rocks/luaffi-scm-1.rockspec
$TORCH_PREFIX/bin/luarocks install $BUILD_ROOT/rocks/cwrap-scm-1.rockspec
$TORCH_PREFIX/bin/luarocks install $BUILD_ROOT/rocks/paths-scm-1.rockspec
#$TORCH_PREFIX/bin/luarocks install $BUILD_ROOT/rocks/torch-scm-1.rockspec
echo "[Pre-build] installing torch7 from source"
cd $BUILD_ROOT
git clone http://github.com/torch/torch7
cd torch7
# patch neon vector itrinsics (this should be fixed in master now)
# cp ../NEON.c lib/TH/vector/NEON.c
# cat lib/TH/vector/NEON.c
# patch set 993
#sed -i '6 a STRING(REGEX REPLACE "^.*(asimd).*$" "\\\\1" ASIMD_THERE ${CPUINFO})' $BUILD_ROOT/torch7/lib/TH/cmake/FindARM.cmake
#sed -i '7 a STRING(COMPARE EQUAL "asimd" "${ASIMD_THERE}" ASIMD_TRUE)' $BUILD_ROOT/torch7/lib/TH/cmake/FindARM.cmake
#sed -i '8 a IF (ASIMD_TRUE)' $BUILD_ROOT/torch7/lib/TH/cmake/FindARM.cmake
#sed -i '9 a set(ASIMD_FOUND true CACHE BOOL "ASIMD/NEON available on host")' $BUILD_ROOT/torch7/lib/TH/cmake/FindARM.cmake
#sed -i '10 a ELSE (ASIMD_TRUE)' $BUILD_ROOT/torch7/lib/TH/cmake/FindARM.cmake
#sed -i '11 a set(ASIMD_FOUND false CACHE BOOL "ASIMD/NEON available on host")' $BUILD_ROOT/torch7/lib/TH/cmake/FindARM.cmake
#sed -i '12 a ENDIF (ASIMD_TRUE)' $BUILD_ROOT/torch7/lib/TH/cmake/FindARM.cmake
#sed -i '64 a IF (ASIMD_FOUND)' $BUILD_ROOT/torch7/lib/TH/CMakeLists.txt
#sed -i '65 a MESSAGE(STATUS "asimd/Neon found with compiler flag : -D__NEON__")' $BUILD_ROOT/torch7/lib/TH/CMakeLists.txt
#sed -i '66 a SET(CMAKE_C_FLAGS "-D__NEON__ ${CMAKE_C_FLAGS}")' $BUILD_ROOT/torch7/lib/TH/CMakeLists.txt
#sed -i '67 a ENDIF (ASIMD_FOUND)' $BUILD_ROOT/torch7/lib/TH/CMakeLists.txt
#cat $BUILD_ROOT/torch7/lib/TH/cmake/FindARM.cmake
#cat $BUILD_ROOT/torch7/lib/TH/CMakeLists.txt
#sed -i 's/#if defined(__arm__)/#if defined(__arm__) || defined(__arm64)/' $BUILD_ROOT/torch7/lib/TH/generic/simd/simd.h
#sed -i 's/#if defined(__arm__)/#if defined(__arm__) || defined(__NEON__)/' $BUILD_ROOT/torch7/lib/TH/generic/simd/simd.h
cat $BUILD_ROOT/torch7/lib/TH/generic/simd/simd.h
$TORCH_PREFIX/bin/luarocks make $BUILD_ROOT/torch7/rocks/torch-scm-1.rockspec
cd $BUILD_ROOT
#$TORCH_PREFIX/bin/luarocks install $BUILD_ROOT/torch7/rocks/torch-scm-1.rockspec
echo "[Pre-build] done installing torch7 package"
echo "[Pre-build] installing additional packages for Torch"
echo "[Pre-build] installing nn from source"
git clone http://github.com/torch/nn
cd nn
#sed -i 's/ptrdiff_t/long/g' lib/THNN/init.c
#sed -i 's/ptrdiff_t/long/g' lib/THNN/generic/*
$TORCH_PREFIX/bin/luarocks make $BUILD_ROOT/nn/rocks/nn-scm-1.rockspec
cd $BUILD_ROOT
echo "[Pre-build] done installing nn package"
#$TORCH_PREFIX/bin/luarocks install $BUILD_ROOT/rocks/nn-scm-1.rockspec
$TORCH_PREFIX/bin/luarocks install $BUILD_ROOT/rocks/nnx-0.1-1.rockspec
$TORCH_PREFIX/bin/luarocks install $BUILD_ROOT/rocks/optim-1.0.5-0.rockspec
$TORCH_PREFIX/bin/luarocks install $BUILD_ROOT/rocks/gnuplot-scm-1.rockspec
$TORCH_PREFIX/bin/luarocks install $BUILD_ROOT/rocks/nngraph-scm-1.rockspec
#$TORCH_PREFIX/bin/luarocks install cwrap
#$TORCH_PREFIX/bin/luarocks install paths
#$TORCH_PREFIX/bin/luarocks install torch
#$TORCH_PREFIX/bin/luarocks install nn
#$TORCH_PREFIX/bin/luarocks install nnx
#$TORCH_PREFIX/bin/luarocks install optim
#$TORCH_PREFIX/bin/luarocks install cutorch
#$TORCH_PREFIX/bin/luarocks install trepl
#$TORCH_PREFIX/bin/luarocks install gnuplot
echo "[Pre-build] installing cutorch from source"
git clone https://github.com/torch/cutorch
sed -i 's/$(getconf _NPROCESSORS_ONLN)/1/g' cutorch/rocks/cutorch-1.0-0.rockspec
sed -i 's/$(getconf _NPROCESSORS_ONLN)/1/g' cutorch/rocks/cutorch-scm-1.rockspec
sed -i 's/jopts=3/jopts=1/g' cutorch/rocks/cutorch-1.0-0.rockspec
sed -i 's/jopts=3/jopts=1/g' cutorch/rocks/cutorch-scm-1.rockspec
$TORCH_PREFIX/bin/luarocks install $BUILD_ROOT/cutorch/rocks/cutorch-scm-1.rockspec
echo "[Pre-build] done installing cutorch package"
echo "[Pre-build] installing cudnn bindings from source"
# install cudnn v5 bindings
#git clone -b R5 http://github.com/soumith/cudnn.torch
git clone http://github.com/soumith/cudnn.torch
#sed -i 's/ffi.sizeof('half'),/2,/g' cudnn.torch/init.lua
$TORCH_PREFIX/bin/luarocks install $BUILD_ROOT/cudnn.torch/cudnn-scm-1.rockspec
echo "[Pre-build] done installing cudnn bindings package"
echo ""
echo "[Pre-build] Torch7 has been installed successfully"
echo ""
#echo "installing iTorch"
#sudo apt-get install libzmq3-dev libssl-dev python-zmq
#sudo pip install ipython
#ipython --version
## pip uninstall IPython
## pip install ipython==3.2.1
#sudo pip install jupyter
#git clone https://github.com/facebook/iTorch.git
#$TORCH_PREFIX/bin/luarocks install $BUILD_ROOT/iTorch/itorch-scm-1.rockspec
fi参考：
https://github.com/dusty-nv/jetson-reinforcement/blob/master/CMakePreBuild.sh
- 本文固定链接: http://rosrobot.cn/?id=258
- 转载请注明: znjrobot 于 北京智能佳科技有限公司 发表
《本文》有 0 条评论