Is it can be run on winodws??When I run "python setup.py install" ,it will show this about neural_renderer

[Zju-George]

@maartenjv Yes that's the problem I mentioned before.
You just need to find rasterize_cuda_kernel.cu, open it with any text editor and find atomicAdd(),
replace the function with

  #if !defined(__CUDA_ARCH__) || __CUDA_ARCH__ >= 600

  #else
  static __inline__ __device__ double atomicAdd(double *address, double val) {
    unsigned long long int* address_as_ull = (unsigned long long int*)address;
    unsigned long long int old = *address_as_ull, assumed;
    if (val==0.0)
      return __longlong_as_double(old);
    do {
      assumed = old;
      old = atomicCAS(address_as_ull, assumed, __double_as_longlong(val +__longlong_as_double(assumed)));
    } while (assumed != old);
    return __longlong_as_double(old);
  }


  #endif

leave a comment if you succeed

[Zju-George]

yes you just find the torch.h file and delete the 7th line,
and you may encounter another problem, then comment again, I will show you how to fix @maartenjv

[BingEdison]

I've compiled using conda , did you compile from pulled git and not pip install of neural renderer?

[nkolot]

There is no support for Windows currently and I don't plan to support it in the future either.

[Zju-George]

after a long time struggling, i build this on windows, if you need you can write Emali:[email protected] to me if you wish

[maartenjv]

Hello, I have the same problem, getting many of these messages during install:

warning C4251: 'torch::optim::RMSprop::grad_average_buffers': class 'std::vector<at::Tensor,std::allocator<_Ty>>' needs to have dll-interface to be used by clients of class 'torch::optim::RMSprop' with [ _Ty=at::Tensor ]

And finally this one:
C:\ProgramData\Anaconda3\lib\site-packages\torch\include\torch\csrc\api\include\torch/torch.h(7): fatal error C1021: invalid preprocessor command 'warning'
error: command 'C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\VC\Tools\MSVC\14.16.27023\bin\HostX86\x64\cl.exe' failed with exit status 2

I am running Windows 7, Anaconda 3, CUDA 10, pytorch 1.1.0

[maartenjv]

Thanks, it seems to be going on for longer now, although the dll warnings are still there.
Now there is a different error:

neural_renderer/cuda/rasterize_cuda_kernel.cu(9): error: function "atomicAdd(double *, double)" has already been defined

1 error detected in the compilation of "....../AppData/Local/Temp/tmpxft_00001ee4_00000000-8_rasterize_cuda_kernel.cpp4.ii".
error: command 'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1\bin\nvcc.exe' failed with exit status 2

[maartenjv]

seems to be working, ill test some more.

thanks a lot!

[BingEdison]

solved issues with windows install , thanks mate!

[v-prgmr]

@BingEdison , hey mate, I created a venv and installed neural-renderer after modifying the rasterize_cuda_kernel.cu and everything installed. However I do get a warning in the early stages of the build - error checking for compiler version, but it proceeds and installs neural-renderer within this venv.

When i try to test one of the examples, I get a ImportError as follows

Traceback (most recent call last):
  File "example4.py", line 15, in <module>
    import neural_renderer as nr
  File "E:\Git Repos\neural_renderer\nr\lib\site-packages\neural_renderer_pytorch-1.1.3-py3.6-win-amd64.egg\neural_renderer\__init__.py", line 3, in <module>
    from .load_obj import load_obj
  File "E:\Git Repos\neural_renderer\nr\lib\site-packages\neural_renderer_pytorch-1.1.3-py3.6-win-amd64.egg\neural_renderer\load_obj.py", line 8, in <module>
    import neural_renderer.cuda.load_textures as load_textures_cuda
ImportError: DLL load failed: The specified module could not be found.

Do you have suggestions?

[v-prgmr]

Yes, I am working on the pulled git. Which branch did you use ? @BingEdison

[yzhao666]

It worked for me. @Zju-George Thanks a million!

[huytuong010101]

@Zju-George Thank u so much, it work for me

[ismeyueyue]

@huytuong010101 Where did you find the torch.h file? I'm struggling on this installation

[Towel2000]

Would anyone please tell me which line of torch.h I actually need to delete?
Mine's has nothing on line7, which is troublesome. I tried installing for more than a month.

[qingqing5r44]

hi, you can find torch related files in your anaconda spin virtual environment. Just search "torch.h" there. Mine is "Anaconda3\envs\spin\Lib\site-packages\torch\include\torch\csrc\api\include\torch\torch.h"

@huytuong010101 Where did you find the torch.h file? I'm struggling on this installation

Hi, you can find torch related files in your anaconda spin virtual environment. Just search "torch.h" there. Mine is "Anaconda3\envs\spin\Lib\site-packages\torch\include\torch\csrc\api\include\torch\torch.h"

[qingqing5r44]

Would anyone please tell me which line of torch.h I actually need to delete? Mine's has nothing on line7, which is troublesome. I tried installing for more than a month.

Hi, you should delete line 7 of torch.h, which is shown below.
#warning \ "Including torch/torch.h for C++ extensions is deprecated. Please include torch/extension.h"

[qingqing5r44]

Thanks to all the instructions above, I have just installed neural_renderer_pytorch successfully in Win10. (using neural_renderer_pytorch-win10 version)

Here I'd like to give one tip we might use while installing. After following the above steps, we may still find errors like this.
neural_renderer/cuda/load_textures_cuda_kernel.cu(103): error: calling a __host__ function("__roundf") from a __global__ function("_NV_ANON_NAMESPACE::load_textures_cuda_kernel<float> ") is not allowed
At this time, you need to change the "floor/ceil/round" into "floorf/ceilf/roundf" in the corresponding source code. Then the errors would disappear.

Hope this notice helps you all :)

[ModeSky]

My environment is windows10, torch1.8, cuda11.3. Running python setup.py install reports the following error. What is the problem