在Nvidia TX2上安装Cuda8.0+Cudnn5.1+TensorFlow1.2+OpenCV3.2

工具

• 宿主机（host）：ubuntu14.04、至少50GB存储空间
• Jetpack3.0：软件下载地址、文档地址

因为只有Jetpack3.0包含cuda8.0和cudnn5.1，而Jetpack3.0只能运行在ubuntu14.04上（亲测16.04有问题会导致失败）

安装Jetpack3.0，并对TX2进行刷机和安装Cuda等框架

保证宿主机和tx2在同一局域网下，按照官方文档和Jetpak的提示执行即可，很简单。

注意事项：

（1）如果在host上安装cuda toolkit出现错误（类似held package等）的时候，换一下ubuntu的更新源（推荐163源）就可以了。

（2）期间如果出现host机提示determining IP …. （卡很久）之类的提示，可以关掉，然后重新来，flash system image那个选择no action，之后就会出现手动配置板子的IP和用户密码的选项。

安装TensorFlow1.2

我编译好的whl文件 —— 百度网盘 提取码：n1sz

可以尝试直接使用我编译好的whl文件（直接跳到 安装whl文件 这一步）

不保证可用，保险起见如果有时间的话最好还是执行下面的步骤。

安装依赖

sudo add-apt-repository ppa:webupd8team/java
sudo apt-get update
sudo apt-get install oracle-java8-installer -y
sudo apt-get install zip unzip autoconf automake libtool curl zlib1g-dev maven -y
sudo apt install python3-numpy python3-dev python3-pip python3-wheel

安装bazel

bazel_version=0.5.1
wget https://github.com/bazelbuild/bazel/releases/download/$bazel_version/bazel-$bazel_version-dist.zip
unzip bazel-$bazel_version-dist.zip -d bazel-dist
sudo chmod -R ug+rwx bazel-dist
cd bazel-dist
#编译这一步很耗时，约20~30分钟
./compile.sh 
sudo cp output/bazel /usr/local/bin

下载TensorFlow源码

cd ~
git clone --recursive https://github.com/tensorflow/tensorflow.git
cd tensorflow
git checkout v1.2.0

修改tensorflow/workspace.bzl,更改为如下：

#注意是eigen_archive节点
native.new_http_archive(
      name = "eigen_archive",urls = [
          "http://mirror.bazel.build/bitbucket.org/eigen/eigen/get/d781c1de9834.tar.gz","https://bitbucket.org/eigen/eigen/get/d781c1de9834.tar.gz",],sha256 = "a34b208da6ec18fa8da963369e166e4a368612c14d956dd2f9d7072904675d9b",strip_prefix = "eigen-eigen-d781c1de9834",build_file = str(Label("//third_party:eigen.BUILD")),)

配置编译变量

export PYTHON_BIN_PATH=$(which python3)
# No Google Cloud Platform support
export TF_NEED_GCP=0
# No Hadoop file system support
export TF_NEED_HDFS=0
# Use CUDA
export TF_NEED_CUDA=1
# Setup gcc ; just use the default
export GCC_HOST_COMPILER_PATH=$(which gcc)
# TF CUDA Version 
export TF_CUDA_VERSION=8.0
# CUDA path
export CUDA_TOOLKIT_PATH=/usr/local/cuda
# cuDNN
export TF_CUDNN_VERSION=5.1.10
export CUDNN_INSTALL_PATH=/usr/lib/aarch64-linux-gnu
# CUDA compute capability
export TF_CUDA_COmpuTE_CAPABILITIES=6.2
export CC_OPT_FLAGS=-march=native
export TF_NEED_JEMALLOC=1
export TF_NEED_OPENCL=0
export TF_ENABLE_XLA=1

编译、生成、打包

#这一步有提示的话，直接enter就行，不用理
./configure
#这一步巨巨巨耗时！约3小时
bazel build -c opt --verbose_failures --config=cuda //tensorflow/tools/pip_package:build_pip_package
bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg
mv /tmp/tensorflow_pkg/tensorflow-1.2.0*-linux_aarch64.whl ~

安装whl文件

#有可能提示版本问题，参考错误提示修改
#这步也很耗时，需要耐心等待
pip3 install ~/tensorflow-1.2.0-cp35-cp35m-linux_aarch64.whl

安装这一步会在“Running setup.py bdist_wheel for numpy …”卡住很久，需要一点耐心

测试脚本testtf.py

#!/usr/bin/env python 

import tensorflow as tf
hello = tf.constant('Hello,TensorFlow!')
sess = tf.Session()
print(sess.run(hello))

运行：

python3 testtf.py

结果：

nvidia@tegra-ubuntu:~$ python3 testtf.py 
2017-12-26 02:30:00.977979: E tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:879] Could not open file to read NUMA node: /sys/bus/pci/devices/0000:00:00.0/numa_node
Your kernel may have been built without NUMA support.
2017-12-26 02:30:00.978096: I tensorflow/core/common_runtime/gpu/gpu_device.cc:940] Found device 0 with properties: 
name: GP10B
major: 6 minor: 2 memoryClockRate (GHz) 1.3005
pciBusID 0000:00:00.0
Total memory: 7.67GiB
Free memory: 3.97GiB
2017-12-26 02:30:00.978144: I tensorflow/core/common_runtime/gpu/gpu_device.cc:961] DMA: 0 
2017-12-26 02:30:00.978174: I tensorflow/core/common_runtime/gpu/gpu_device.cc:971] 0:   Y 
2017-12-26 02:30:00.978204: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1030] Creating TensorFlow device (/gpu:0) -> (device: 0,name: GP10B,pci bus id: 0000:00:00.0)
2017-12-26 02:30:00.978237: I tensorflow/core/common_runtime/gpu/gpu_device.cc:642] Could not identify NUMA node of /job:localhost/replica:0/task:0/gpu:0,defaulting to 0.  Your kernel may not have been built with NUMA support.
2017-12-26 02:30:02.406679: I tensorflow/compiler/xla/service/platform_util.cc:58] platform CUDA present with 1 visible devices
2017-12-26 02:30:02.406746: I tensorflow/compiler/xla/service/platform_util.cc:58] platform Host present with 4 visible devices
2017-12-26 02:30:02.407489: I tensorflow/compiler/xla/service/service.cc:198] XLA service 0x29c9470 executing computations on platform Host. Devices:
2017-12-26 02:30:02.407540: I tensorflow/compiler/xla/service/service.cc:206]   StreamExecutor device (0): <undefined>,<undefined>
2017-12-26 02:30:02.408398: I tensorflow/compiler/xla/service/platform_util.cc:58] platform CUDA present with 1 visible devices
2017-12-26 02:30:02.408446: I tensorflow/compiler/xla/service/platform_util.cc:58] platform Host present with 4 visible devices
2017-12-26 02:30:02.409154: I tensorflow/compiler/xla/service/service.cc:198] XLA service 0x2a193b0 executing computations on platform CUDA. Devices:
2017-12-26 02:30:02.409199: I tensorflow/compiler/xla/service/service.cc:206]   StreamExecutor device (0): GP10B,Compute Capability 6.2
b'Hello,TensorFlow!'

OpenCV3.2

参考文献：https://docs.opencv.org/3.2.0/d6/d15/tutorial_building_tegra_cuda.html

下载opencv3.2源码

wget https://github.com/opencv/opencv/archive/3.2.0.zip
unzip 3.2.0.zip
cd opencv-3.2.0/

安装依赖项

sudo apt-get install \
    cmake \
    libglew-dev \
    libtiff5-dev \
    zlib1g-dev \
    libjpeg-dev \
    libpng12-dev \
    libjasper-dev \
    libavcodec-dev \
    libavformat-dev \
    libavutil-dev \
    libpostproc-dev \
    libswscale-dev \
    libeigen3-dev \
    libtbb-dev \
    libgtk2.0-dev \
    pkg-config

配置、编译并安装

mkdir build 
cd build
cmake \
    -DCMAKE_BUILD_TYPE=Release \
    -DCMAKE_INSTALL_PREFIX=/usr \
    -DBUILD_PNG=OFF \
    -DBUILD_TIFF=OFF \
    -DBUILD_TBB=OFF \
    -DBUILD_JPEG=OFF \
    -DBUILD_JASPER=OFF \
    -DBUILD_ZLIB=OFF \
    -DBUILD_EXAMPLES=ON \
    -DBUILD_opencv_java=OFF \
    -DBUILD_opencv_python2=ON \
    -DBUILD_opencv_python3=ON \
    -DBUILD_PYTHON_SUPPORT=ON \
    -DENABLE_precompiled_HEADERS=OFF \
    -DWITH_OPENCL=OFF \
    -DWITH_OPENMP=OFF \
    -DWITH_FFMPEG=ON \
    -DWITH_GSTREAMER=OFF \
    -DWITH_GSTREAMER_0_10=OFF \
    -DWITH_CUDA=ON \
    -DWITH_GTK=ON \
    -DWITH_VTK=OFF \
    -DWITH_TBB=ON \
    -DWITH_1394=OFF \
    -DWITH_openexr=OFF \
    -DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-8.0 \
    -DCUDA_ARCH_BIN=6.2 \
    -DCUDA_ARCH_PTX="" \
    -DINSTALL_C_EXAMPLES=ON \
    -DINSTALL_TESTS=OFF \
    -DOPENCV_TEST_DATA_PATH=../opencv_extra/testdata \
    ..
#编译需要一个小时左右
make -j4
sudo make install

作者：yanjie 2017.12.26