（原）ubuntu16重装显卡驱动后，torch中的问题

转载请注明出处：

http://www.cnblogs.com/darkknightzh/p/6030232.html

参考网址：

https://github.com/torch/cutorch/issues/210

https://github.com/torch/cutorch/issues/425

https://developer.nvidia.com/cuda-downloads

http://tieba.baidu.com/p/4210823766

https://github.com/torch/cutorch

https://github.com/torch/cutorch

https://github.com/torch/cutorch/issues/480

https://github.com/torch/cunn

https://github.com/soumith/cudnn.torch/issues/279

说明：最好直接安装lua5.2，而后torch，cutorch，cunn等都安装新版本的，否则可能有冲突。

前几天重新安装了最新的显卡驱动，结果今天使用torch的require ‘cutorch’时，报错（类似下面的信息）：

cannot load torch/install/lib/lua/5.1/libcutorch.so

https://github.com/torch/cutorch/issues/210说要重新安装cutorch，使用luarocks install cutorch后，又提示Specify CUDA_TOOLKIT_ROOT_DIR，如下：

网上一搜，https://github.com/torch/cutorch/issues/425说是没有安装cuda。。。

1. 由于之前重装显卡驱动时，已经将cuda删除了。因而此时需要重新安装cuda。索性安装cuda8.0。官网https://developer.nvidia.com/cuda-downloads下载ubuntu16.04的版本。之后按照官网说明，在终端中输入：

sudo dpkg -i cuda-repo-ubuntu1604---local_8.0.44-1_amd64.deb
sudo apt-get update
sudo apt-get install cuda

2. 此时直接在终端中输入nvcc --version时，还是会提示找不到nvcc。http://tieba.baidu.com/p/4210823766中9楼指出，需要添加环境变量。在终端中输入gedit .bashrc，最后添加（最后一个不确定是否需要添加）：

export PATH=/usr/local/cuda-8.0/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-8.0/lib64:$LD_LIBRARY_PATH
export CUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-8.0:$CUDA_TOOLKIT_ROOT_DIR

重启后，输入nvcc --version，成功：

3. 下载cutorch，网址：https://github.com/torch/cutorch

4. 终端定位到该文件夹内，而后输入：

luarocks install rocks/cutorch-1.0-.rockspec

等待安装完成。

说明：rocks文件夹内有cutorch-1.0-0.rockspec和cutorch-scm-1.rockspec两个文件。如果使用如下命令：

luarocks install rocks/cutorch-scm-1.rockspec

会提示：

1 error detected in the compilation of "/tmp/tmpxft_00000c32_00000000-7_THCTensorMath.cpp1.ii".
CMake Error at THC_generated_THCTensorMath.cu.o.cmake:267 (message):
  Error generating file
  /tmp/luarocks_cutorch-scm-1-1850/cutorch/build/lib/THC/CMakeFiles/THC.dir//./THC_generated_THCTensorMath.cu.o

lib/THC/CMakeFiles/THC.dir/build.make:105: recipe for target 'lib/THC/CMakeFiles/THC.dir/THC_generated_THCTensorMath.cu.o' failed
make[2]: *** [lib/THC/CMakeFiles/THC.dir/THC_generated_THCTensorMath.cu.o] Error 1
make[2]: *** 正在等待未完成的任务....
/tmp/luarocks_cutorch-scm-1-1850/cutorch/lib/THC/THCTensorMathReduce.cuh(471): error: identifier "TH_INDEX_BASE" is undefined

/tmp/luarocks_cutorch-scm-1-1850/cutorch/lib/THC/THCTensorMathReduce.cuh(552): error: identifier "TH_INDEX_BASE" is undefined

2 errors detected in the compilation of "/tmp/tmpxft_00000c45_00000000-5_THCTensorMath2.cpp4.ii".
CMake Error at THC_generated_THCTensorMath2.cu.o.cmake:267 (message):
  Error generating file
  /tmp/luarocks_cutorch-scm-1-1850/cutorch/build/lib/THC/CMakeFiles/THC.dir//./THC_generated_THCTensorMath2.cu.o

lib/THC/CMakeFiles/THC.dir/build.make:112: recipe for target 'lib/THC/CMakeFiles/THC.dir/THC_generated_THCTensorMath2.cu.o' failed
make[2]: *** [lib/THC/CMakeFiles/THC.dir/THC_generated_THCTensorMath2.cu.o] Error 1
/tmp/luarocks_cutorch-scm-1-1850/cutorch/lib/THC/THCTensorMathReduce.cuh(471): error: identifier "TH_INDEX_BASE" is undefined

/tmp/luarocks_cutorch-scm-1-1850/cutorch/lib/THC/THCTensorMathReduce.cuh(552): error: identifier "TH_INDEX_BASE" is undefined

2 errors detected in the compilation of "/tmp/tmpxft_00000cd5_00000000-5_THCTensorMathScan.cpp4.ii".
CMake Error at THC_generated_THCTensorMathScan.cu.o.cmake:267 (message):
  Error generating file
  /tmp/luarocks_cutorch-scm-1-1850/cutorch/build/lib/THC/CMakeFiles/THC.dir//./THC_generated_THCTensorMathScan.cu.o

lib/THC/CMakeFiles/THC.dir/build.make:147: recipe for target 'lib/THC/CMakeFiles/THC.dir/THC_generated_THCTensorMathScan.cu.o' failed
make[2]: *** [lib/THC/CMakeFiles/THC.dir/THC_generated_THCTensorMathScan.cu.o] Error 1
/tmp/luarocks_cutorch-scm-1-1850/cutorch/lib/THC/THCTensorMathReduce.cuh(471): error: identifier "TH_INDEX_BASE" is undefined

/tmp/luarocks_cutorch-scm-1-1850/cutorch/lib/THC/THCTensorMathReduce.cuh(552): error: identifier "TH_INDEX_BASE" is undefined

2 errors detected in the compilation of "/tmp/tmpxft_00000cc9_00000000-5_THCTensorMathReduce.cpp4.ii".
CMake Error at THC_generated_THCTensorMathReduce.cu.o.cmake:267 (message):
  Error generating file
  /tmp/luarocks_cutorch-scm-1-1850/cutorch/build/lib/THC/CMakeFiles/THC.dir//./THC_generated_THCTensorMathReduce.cu.o

lib/THC/CMakeFiles/THC.dir/build.make:140: recipe for target 'lib/THC/CMakeFiles/THC.dir/THC_generated_THCTensorMathReduce.cu.o' failed
make[2]: *** [lib/THC/CMakeFiles/THC.dir/THC_generated_THCTensorMathReduce.cu.o] Error 1
CMakeFiles/Makefile2:172: recipe for target 'lib/THC/CMakeFiles/THC.dir/all' failed
make[1]: *** [lib/THC/CMakeFiles/THC.dir/all] Error 2
Makefile:127: recipe for target 'all' failed
make: *** [all] Error 2

Error: Build error: Failed building.

https://github.com/torch/cutorch/issues/480指出，需要更新torch，nn，cunn等。懒得更新了，因而就使用老版本的了。

5. 此时再require ‘cutorch’便不会出错。

6. 此时若要使用cunn，还是会报错：

在官网下载cunn：https://github.com/torch/cunn，并安装cunn-1.0-0.rockspec：

luarocks install cunn-1.0-.rockspec

此处若安装新的cunn-scm-1.rockspec，也会报错：

安装完之后，require ‘cunn’成功。

7. 之后require ‘cudnn’再次遇到问题。重装之后，require时，会提示：

attempt to call field 'hasFastHalfInstructions' (a nil value)

https://github.com/soumith/cudnn.torch/issues/279指出，需要更新lua到5.2。

暂时不使用cndnn了，不更新了。重新配置起来太麻烦了。。。