TypeError when installing from source on windows, https://discuss.pytorch.org/

1

I’ve encountered problem when installing pytorch from source (the ‘master’ branch) in ‘Anaconda’ following the instruction . I’m interested in installing from source because i attempt to allow pyTorch to support my old GPU.

Can anyone help to explain why it needs to set for the ‘NVTOOLEXT_HOME’ env variable ? anything might be potentially missing in my installation steps ?

Thanks,

...
(pytorchtest) C:\myproject\python>call "C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\VC\Auxiliary\Build\vcvarsall.bat" x64 -vcvars_ver=14.11
**********************************************************************
** Visual Studio 2017 Developer Command Prompt v15.8.3
** Copyright (c) 2017 Microsoft Corporation
**********************************************************************
[ERROR:vcvars.bat] Toolset directory for version '14.11' was not found.
[ERROR:VsDevCmd.bat] *** VsDevCmd.bat encountered errors. Environment may be incomplete and/or incorrect. ***
[ERROR:VsDevCmd.bat] In an uninitialized command prompt, please 'set VSCMD_DEBUG=[value]' and then re-run
[ERROR:VsDevCmd.bat] vsdevcmd.bat [args] for additional details.
[ERROR:VsDevCmd.bat] Where [value] is:
[ERROR:VsDevCmd.bat]    1 : basic debug logging
[ERROR:VsDevCmd.bat]    2 : detailed debug logging
[ERROR:VsDevCmd.bat]    3 : trace level logging. Redirection of output to a file when using this level is recommended.
[ERROR:VsDevCmd.bat] Example: set VSCMD_DEBUG=3
[ERROR:VsDevCmd.bat]          vsdevcmd.bat > vsdevcmd.trace.txt 2>&1
Building wheel torch-0.5.0a0+70d93f4
Traceback (most recent call last):
  File "C:\myproject\python\pytorch\setup.py", line 919, in <module>
    nvtoolext_lib_path = NVTOOLEXT_HOME + '/lib/x64/'
TypeError: unsupported operand type(s) for +: 'NoneType' and 'str'
2

There is a problem when installing CUDA 9.2 with VS integration. I had to deselect VS integration to finish CUDA installation. It turns out that NAVIDIA NSight is looking for ‘devenv.com’ from different path. So, it is a problem when installing Microsoft Visual Studio community 2017 edition. Reinstalling does not work. After applying the workaround [1], NVTX can be installed successfully.

However, when attempting to install from source again, it complains:

c:\program files\nvidia gpu computing toolkit\cuda\v9.2\include\crt/host_config.h(133): fatal error C1189: #error:  -- unsupported Microsoft Visual Studio version! Only the versions 2012, 2013, 2015 and 2017 are supported! [C:\oak-project\python\pytorch\build\caffe2\caffe2_gpu.vcxproj]

I only have VS 2017 installed and set ‘-vcvars_ver=14.15’ (by checking the actual directory in C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\VC\Tools\MSVC)

[1] https://devtalk.nvidia.com/default/topic/1016018/nsight-visual-studio-edition/nsight-installation-problem-windows-10-setup-wizard-ended-prematurely-because-of-an-error/

3

Build can continue by manually changing line 131 to #if _MSC_VER < 1600 || _MSC_VER > 1915

see also: https://stackoverflow.com/questions/47645436/cuda-9-unsupported-error-with-vs-2017

4

now, i’m stuck in compiling caffe2. It complains:

  C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\Common7\IDE\VC\VCTargets\Microsoft.CppCommon.targets(209,5): error MSB6006: "cmd.exe" exited with code 1. [C:\python\pytorch\build\caffe2\caffe2_gpu.vcxproj]
5

cc @peterjc123 do you have any idea?

6

Would you please give me the full log? This error message is not the root cause of the build failure.

7

I’m afraid that you are using the wrong version of VS. I have not tested VS 14.15 before but actually you can try to install 14.11. That’s the version we used in the build.