tensorflow 源码编译遇到的坑坑坑坑坑坑坑坑

前言

　强烈推荐cuda8.0+cudnn5+ 然后直接pip安装！源码编译非常费劲　很多坑　de的我都意识模糊了

配置：ubuntu14,cuda7.5,cudnn版本未知(应该是5吧)
安装tensorflow 1.0.1-gpu
需要准备一台能疯狂google github issues的电脑

因为是cuda7.5还是公用服务器。被迫源码编译

进入主题

　clone 要从github下载zip文件手动解压，不要用git clone, 否则丢文件git check 不出来但是编译会报错【原因未知】
　
　大体按照官网安装。也就是install from source

　装bazel.　bazel 官网install页面挂了，去百度搜别人博客基本能装上。

$ cd tensorflow  # cd to the top-level directory created
$ ./configure

　这两行开始问题就比较大
　首先 ./configure 不要用sudo!, 如果没权限就把tensorflow目录cp 到能不用sudo的

XLA just-in-time compiler (experimental)? [y/N]　

　点Ｎ

所有让输地址的尽量不要用default.挨个查清楚装在哪了再输进去

　如果./configure报错，那就重新输，直到成功或者报错信息连续多次一样为止，因为有时候只是网不好

　Please note that each additional compute capability significantly increases your build time and binary size.　[Default is: "3.5,5.2"]: 6.1

　这个要查一下，titan所以6.1,　令人惊奇的是我的笔记本居然是5.0

大坑

$ bazel build --config=opt --config=cuda //tensorflow/tools/pip_package:build_pip_package

这行。会报错

ERROR: /home/darth/.cache/bazel/_bazel_darth/08554d152596e5a7df399506682a63f3/external/local_config_cuda/crosstool/BUILD:4:1: Traceback (most recent call last):
    File "/home/darth/.cache/bazel/_bazel_darth/08554d152596e5a7df399506682a63f3/external/local_config_cuda/crosstool/BUILD", line 4
        error_gpu_disabled()
    File "/home/darth/.cache/bazel/_bazel_darth/08554d152596e5a7df399506682a63f3/external/local_config_cuda/crosstool/error_gpu_disabled.bzl", line 3, in error_gpu_disabled
        fail("ERROR: Building with --config=c...")
ERROR: Building with --config=cuda but TensorFlow is not configured to build with GPU support. Please re-run ./configure and enter 'Y' at the prompt to build with GPU support.
ERROR: no such target '@local_config_cuda//crosstool:toolchain': target 'toolchain' not declared in package 'crosstool' defined by /home/darth/.cache/bazel/_bazel_darth/08554d152596e5a7df399506682a63f3/external/local_config_cuda/crosstool/BUILD.
INFO: Elapsed time: 0.097s

划重点

 '@local_config_cuda//crosstool:toolchain': target 'toolchain' not declared in package 'crosstool' defined by 

　解决：更新.bashrc里面的cuda变量
Problem on building target with GPU support #4841
　
　也就是

export "TF_NEED_CUDA=1" in my ~/.bashrc

　我是用vim 写进去的，别忘了出来source更新。