CODEWITHSUNDEEP
tensorflowmsbuildcmakewindows-10
HelloGoodbye
HelloGoodbye

Reputation: 3902

Errors when building TensorFlow with CMake on Windows 10

I am attempting a TensorFlow CMake build on Windows, but I'm running into problems.

First, I run

C:\work\tensorflow\tensorflow\contrib\cmake\build>cmake .. -A x64 -DCMAKE_BUILD_TYPE=Release ^
More? -DSWIG_EXECUTABLE=C:/Program1/swigwin-3.0.12/swig.exe ^
More? -DPYTHON_EXECUTABLE=C:/Python3.5/python.exe ^
More? -DPYTHON_LIBRARIES=C:/Python3.5/libs/python35.lib ^
More? -Dtensorflow_ENABLE_GPU=ON ^
More? -DCUDNN_HOME="C:\Program1\cudnn\cuda"

and get the output

-- Building for: Visual Studio 14 2015
-- The C compiler identification is MSVC 19.0.24215.1
-- The CXX compiler identification is MSVC 19.0.24215.1
-- Check for working C compiler: C:/Program Files (x86)/Microsoft Visual Studio 14.0/VC/bin/x86_amd64/cl.exe
-- Check for working C compiler: C:/Program Files (x86)/Microsoft Visual Studio 14.0/VC/bin/x86_amd64/cl.exe -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working CXX compiler: C:/Program Files (x86)/Microsoft Visual Studio 14.0/VC/bin/x86_amd64/cl.exe
-- Check for working CXX compiler: C:/Program Files (x86)/Microsoft Visual Studio 14.0/VC/bin/x86_amd64/cl.exe -- works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Performing Test COMPILER_OPT_ARCH_NATIVE_SUPPORTED
-- Performing Test COMPILER_OPT_ARCH_NATIVE_SUPPORTED - Failed
-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v8.0 (found suitable version "8.0", minimum required is "8.0")
-- Found PythonInterp: C:/Python3.5/python.exe (found version "3.5")
-- Found PythonLibs: C:/Python3.5/libs/python35.lib (found version "3.5.0")
-- Found SWIG: C:/Program1/swigwin-3.0.12/swig.exe (found version "3.0.12")
-- Configuring done
-- Generating done
-- Build files have been written to: C:/work/tensorflow/tensorflow/contrib/cmake/build

The only troubling line here is -- Performing Test COMPILER_OPT_ARCH_NATIVE_SUPPORTED - Failed, which I don't really know what it means or whether it's a serious problem. (Is it?)

Second, I run

MSBuild /p:Configuration=Release tf_tutorials_example_trainer.vcxproj

and this gives me several errors:

"C:\work\tensorflow\tensorflow\contrib\cmake\build\tf_tutorials_example_trainer.vcxproj" (default target) (1) ->
"C:\work\tensorflow\tensorflow\contrib\cmake\build\tf_core_gpu_kernels.vcxproj" (default target) (90) ->
(CustomBuild target) ->
  C:\Program Files (x86)\MSBuild\Microsoft.Cpp\v4.0\V140\Microsoft.CppCommon.targets(171,5): error MSB6006: "cmd.exe" exited with code 1. [C:\work\tensorflow\tensorflow\contrib\cmake\build\tf_core_gpu_kernels.vcxproj]


"C:\work\tensorflow\tensorflow\contrib\cmake\build\tf_tutorials_example_trainer.vcxproj" (default target) (1) ->
"C:\work\tensorflow\tensorflow\contrib\cmake\build\tf_core_kernels.vcxproj" (default target) (91) ->
(ClCompile target) ->
  c:\work\tensorflow\tensorflow\core\kernels\cuda_solvers.h(24): fatal error C1083: Cannot open include file: 'cuda/include/cusolverDn.h': No such file or directory (compiling source file C:\work\tensorflow\tensorflow\core\kernels\cholesky_op.cc) [C:\work\tensorflow\tensorflow\contrib\cmake\build\tf_core_kernels.vcxproj]
  c:\work\tensorflow\tensorflow\core\kernels\cuda_solvers.h(24): fatal error C1083: Cannot open include file: 'cuda/include/cusolverDn.h': No such file or directory (compiling source file C:\work\tensorflow\tensorflow\core\kernels\cuda_solvers.cc) [C:\work\tensorflow\tensorflow\contrib\cmake\build\tf_core_kernels.vcxproj]
  c:\work\tensorflow\tensorflow\contrib\cmake\build\external\eigen_archive\eigen\src\core\coreevaluators.h(1052): fatal error C1060: compiler is out of heap space (compiling source file C:\work\tensorflow\tensorflow\core\kernels\svd_op_float.cc) [C:\work\tensorflow\tensorflow\contrib\cmake\build\tf_core_kernels.vcxproj]

    696 Warning(s)
    4 Error(s)

I don't know relative to which folder MSBuild is looking for cuda/include/cusolverDn.h. I don't have any folder called cuda in c:\work\tensorflow\tensorflow\core\kernels, and no file named cusolverDn.h exists anywhere in the tensorflow repository. (I do have the file cusolverDn.h in my CUDA install directory, though.) So,

What can I do to make MSBuild find cusolverDn.h? And what can I do to prevent the compiler from running out of heap space?

In a previous attempt to run the last command I also received the message

nvcc fatal   : Microsoft Visual Studio configuration file 'vcvars64.bat' could not be found for installation at 'C:/Program Files (x86)/Microsoft Visual Studio 14.0/VC/bin/../..'

Some recommend renaming a folder and a file in that folder. I can't really verify that this works right now since the problem temporarily seems to have disappeared (maybe masked by the "compiler is out of heap space" problem, which I didn't have before...).

Is renaming this file and this folder the correct solution?

Upvotes: 1

Views: 1888

Answers (1)

Luca Cappa
Luca Cappa

Reputation: 1999

As per tensorflow instructions check the following:

  • you must be running a 64 bit enabled Command Prompt for Visual Studio; you should run your msbuild.exe commands from such a prompt;
  • install CUDNN's Nvidia zio file and append its /bin/ directory to your PATH environment variable, it should contains two entries similar to the following:
    • D:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\bin
    • D:\YOURCUDNNPATH\cuda\bin
  • check whether both git and cmake command are available in the PATH; check whether you could run those commands from your prompt;

Upvotes: 1

Related Questions