TF v 1.7.0 on tx2

[q1178847289]

Thank you very much for sharing resources, but I encountered some problems in the process of compiling tensorflow. I hope you can help me to solve it. Thank you very much! code This bash script will install TensorFlow
branch on a Jetson system that has been setup
by Jetpack with CUDA and cuDNN already installed.

If this is not the case then this script will
likely fail

Expect this script to take up to 6+ hours

Writen by: Jason Tichy < jtichy@nvidia.com >
Version 1.0: Jan 3rd, 2018
Version 1.1: Mar 30, 2018 Added TensorRT support

Note: TF v 1.7.0 release contains a bug for arm because of a hardcoded x86 path in the TensorRT Bazel script, you will need to use master to
build with TensorRT support
Get:1 file:/var/cuda-repo-9-0-local InRelease Ign:1 file:/var/cuda-repo-9-0-local InRelease Get:2 file:/var/nv-tensorrt-repo-ga-cuda9.0-trt3.0.4-20180208 InRelease Ign:2 file:/var/nv-tensorrt-repo-ga-cuda9.0-trt3.0.4-20180208 InRelease Get:3 file:/var/visionworks-repo InRelease Ign:3 file:/var/visionworks-repo InRelease Get:4 file:/var/visionworks-sfm-repo InRelease Ign:4 file:/var/visionworks-sfm-repo InRelease Get:5 file:/var/visionworks-tracking-repo InRelease Ign:5 file:/var/visionworks-tracking-repo InRelease Get:6 file:/var/cuda-repo-9-0-local Release [574 B] Get:7 file:/var/nv-tensorrt-repo-ga-cuda9.0-trt3.0.4-20180208 Release [574 B] Get:8 file:/var/visionworks-repo Release [1,999 B] Get:9 file:/var/visionworks-sfm-repo Release [2,003 B] Get:6 file:/var/cuda-repo-9-0-local Release [574 B] Get:10 file:/var/visionworks-tracking-repo Release [2,008 B] Get:7 file:/var/nv-tensorrt-repo-ga-cuda9.0-trt3.0.4-20180208 Release [574 B] Get:8 file:/var/visionworks-repo Release [1,999 B]
Get:9 file:/var/visionworks-sfm-repo Release [2,003 B]
Get:10 file:/var/visionworks-tracking-repo Release [2,008 B]
Hit:11 http://mirrors.tuna.tsinghua.edu.cn/ubuntu-ports xenial-updates InRelease
Hit:12 http://mirrors.tuna.tsinghua.edu.cn/ubuntu-ports xenial-security InRelease Hit:13 http://mirrors.tuna.tsinghua.edu.cn/ubuntu-ports xenial-backports InRelease Hit:14 http://mirrors.tuna.tsinghua.edu.cn/ubuntu-ports xenial InRelease Hit:18 http://ppa.launchpad.net/webupd8team/java/ubuntu xenial InRelease
Reading package lists... Done
Reading package lists... Done Building dependency tree
Reading state information... Done htop is already the newest version (2.0.1-1ubuntu1). mlocate is already the newest version (0.26-1ubuntu2). ncdu is already the newest version (1.11-1build1). The following packages were automatically installed and are no longer required: apt-clone archdetect-deb dmeventd dmraid dpkg-repack gir1.2-timezonemap-1.0 gir1.2-xkl-1.0 kpartx kpartx-boot libappstream3 libdebian-installer4 libdevmapper-event1.02.1 libdmraid1.0.0.rc16 liblockfile-bin liblockfile1 liblvm2app2.2 liblvm2cmd2.02 libparted-fs-resize0 libreadline5 lockfile-progs lvm2 os-prober pmount python3-icu python3-pam rdate ubiquity-casper ubiquity-ubuntu-artwork ubuntu-core-launcher Use 'sudo apt autoremove' to remove them. 0 upgraded, 0 newly installed, 0 to remove and 28 not upgraded. Build label: 0.10.0- (@non-git) Build target: bazel-out/arm-opt/bin/src/main/java/com/google/devtools/build/lib/bazel/BazelServer_deploy.jar Build time: Tue May 14 13:01:18 +50385 (1527841400478) Build timestamp: 1527841400478 Build timestamp as int: 1527841400478 M tensorflow/contrib/lite/kernels/internal/BUILD M third_party/png.BUILD Already on 'master' Your branch is up-to-date with 'origin/master'. error: patch failed: tensorflow/contrib/lite/kernels/internal/BUILD:20 error: tensorflow/contrib/lite/kernels/internal/BUILD: patch does not apply error: patch failed: third_party/png.BUILD:35 error: third_party/png.BUILD: patch does not apply nvidia@tegra-ubuntu:~/JetsonTFBuild-master$ bazel clean

[zkhodzhaev]

I have exactly the same problem. The first time I was installing, it went well but I was out of space. The next time I tried to install I got the same error.

Any idea how to fix it ?

[idavis]

@Zulfidin @JasonAtNvidia I have a fork to do this but I have to add a couple of things which I didn't want to recommend mainline without @JasonAtNvidia taking a look. The TensorRT install path doesn't work on my PX2 so I have to hard code it.

[JasonAtNvidia]

I'm taking a look at the script soon and will be adding more automation pieces. @idavis Your changes look good. I will probably use them as inspiration as I add a few more automation pieces into the code. I hope to do a rebase soon with something more robust to failure. The original version of this script wasn't a script at all but just a list of how-to, that list became a bash script, that script has grown to include automation. Tensorflow keeps changing things at a rate I cannot keep up with.

[idavis]

@JasonAtNvidia Agreed. It is hard to keep up and figure out which versions of bazel, which defines, etc keep changing. Have you tried cross-compiling TF for Jetson?

[JasonAtNvidia]

@idavis It is my understanding that I cannot cross compile because I would need the ARM source for CUDA, cuDNN, and TensorRT on the computer I am trying to cross compile for. The 4-5 hour compile time on the Jetson itself isn't breaking me down to a point where I am trying to do it on my big workstation yet.