Search code examples
python-3.xwindowstensorflowvisual-studio-2017cmake-gui

Compile tensorflow and error : initializer not allowed for __shared__ variable

I tried to compile tensorflow-1.7.0 in my computer and software included

cuda-10.0

cudnn-7.3

python-3.6.2,

visual studio 2017

cmake.

There are errors :

I don't know how it occurs and how to solve it.

：“C:\Users\12711\Downloads\tensorflow\tensorflow\contrib\cmake\build\tf_python_build_pip_package.vcxproj”(默认目标) (1) ->
“C:\Users\12711\Downloads\tensorflow\tensorflow\contrib\cmake\build\pywrap_tensorflow_internal.vcxproj”(默认目标) (2) ->
“C:\Users\12711\Downloads\tensorflow\tensorflow\contrib\cmake\build\tf_core_gpu_kernels.vcxproj”(默认目标) (138) ->
(CustomBuild 目标) ->
  C:/Users/12711/Downloads/tensorflow\tensorflow/core/kernels/reduction_gpu_kernels.cu.h(271): error : initializer not
allowed for __shared__ variable [C:\Users\12711\Downloads\tensorflow\tensorflow\contrib\cmake\build\tf_core_gpu_kernels
.vcxproj]
  C:/Users/12711/Downloads/tensorflow\tensorflow/core/kernels/reduction_gpu_kernels.cu.h(319): error : initializer not
allowed for __shared__ variable [C:\Users\12711\Downloads\tensorflow\tensorflow\contrib\cmake\build\tf_core_gpu_kernels
.vcxproj]
  C:/Users/12711/Downloads/tensorflow\tensorflow/core/kernels/reduction_gpu_kernels.cu.h(271): error : initializer not
allowed for __shared__ variable [C:\Users\12711\Downloads\tensorflow\tensorflow\contrib\cmake\build\tf_core_gpu_kernels
.vcxproj]
  C:/Users/12711/Downloads/tensorflow\tensorflow/core/kernels/reduction_gpu_kernels.cu.h(319): error : initializer not
allowed for __shared__ variable [C:\Users\12711\Downloads\tensorflow\tensorflow\contrib\cmake\build\tf_core_gpu_kernels
.vcxproj]
  C:/Users/12711/Downloads/tensorflow\tensorflow/core/kernels/reduction_gpu_kernels.cu.h(271): error : initializer not
allowed for __shared__ variable [C:\Users\12711\Downloads\tensorflow\tensorflow\contrib\cmake\build\tf_core_gpu_kernels
.vcxproj]
  C:/Users/12711/Downloads/tensorflow\tensorflow/core/kernels/reduction_gpu_kernels.cu.h(319): error : initializer not
allowed for __shared__ variable [C:\Users\12711\Downloads\tensorflow\tensorflow\contrib\cmake\build\tf_core_gpu_kernels
.vcxproj]
Solution

  • I believe TensorFlow does not support CUDA 10/Windows as yet - confirmed by absence of precompiled binaries for CUDA 10 anywhere online. Recompiled using CUDA 9 and everything worked smoothly.