build cherrypi on windows : python setup.py build command fail

[liongis]

-- Caffe2: Header version is: 9.0
-- Found CUDNN: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v9.0/include

-- Found cuDNN: v7.4.2 (include: C:/Program Files/NVIDIA GPU Computing Toolkit/
CUDA/v9.0/include, library: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v
9.0/lib/x64/cudnn.lib)
CMake Error at cmake/public/cuda.cmake:278 (message):
CUDA support not available with 32-bit windows. Did you forget to set
Win64 in the generator target?
Call Stack (most recent call first):
cmake/Dependencies.cmake:626 (include)
CMakeLists.txt:200 (include)

-- Configuring incomplete, errors occurred!
See also "E:/Game/BWAPI/TorchCraftAI/3rdparty/pytorch/build/CMakeFiles/CMakeOutp
ut.log".
See also "E:/Game/BWAPI/TorchCraftAI/3rdparty/pytorch/build/CMakeFiles/CMakeErro
r.log".

E:\Game\BWAPI\TorchCraftAI\3rdparty\pytorch\build>IF ERRORLEVEL 1 exit 1
Failed to run 'tools\build_pytorch_libs.bat --use-cuda --use-fbgemm --use-nnpack
--use-mkldnn --use-qnnpack caffe2'

[ebetica]

It sounds like you forgot to run

set CMAKE_GENERATOR=Visual Studio 15 2017 Win64

before you started building, as mentioned in Initialize Environment

[liongis]

My compilation environment is windows7,CUDA Toolkit 9.0, python 37-32

5>Building NVCC (Device) object caffe2/CMakeFiles/caffe2_gpu.dir//aten/src/THC/Release/caffe2_gpu_generated_THCReduceApplyUtils.cu.obj
5>nvcc fatal : Unsupported gpu architecture 'compute_20'
5>CMake Error at caffe2_gpu_generated_THCReduceApplyUtils.cu.obj.Release.cmake:219 (message):
5> Error generating
5> E:/Game/BWAPI/TorchCraftAI/3rdparty/pytorch/build/caffe2/CMakeFiles/caffe2_gpu.dir//aten/src/THC/Release/caffe2_gpu_generated_THCReduceApplyUtils.cu.obj
5>
5>
5>D:\Program Files\Microsoft Visual Studio\2017\Community\Common7\IDE\VC\VCTargets\Microsoft.CppCommon.targets(209,5): error MSB6006: “cmd.exe”已退出，代码为 1。

[ebetica]

Which GPU are you using? From a Google search, it's seems that compute capability 2.0 was removed in cuda 9, and if you're using a sufficiently old GPU you might need CUDA 8: kaldi-asr/kaldi#1918

[liongis]

I see how this works

Delete "- gencode arch = compute_20, code = sm_20"

But I don't know in what file to delete this statement.
Is this statement configured in this project or where?

[ebetica]

I'm not sure, we rely on PyTorch to build their system correctly, you might find more information asking in their github repository. https://github.com/pytorch/pytorch/