Installation on windows. RuntimeError: Not compiled with GPU support

[wwdok]

Although the Issues alrady have some similar issues, but i didn't notice a workable solution, so i open a new issue.

My computer system is Windows 10 .
Pytorch env are :

I install pytorch3d by :

git clone https://github.com/facebookresearch/pytorch3d.git
cd pytorch3d && pip install -e .

then i use pytorch3d to run this repo: https://github.com/ascust/3DMM-Fitting-Pytorch, it reports error:

PS D:\AI\3DMM\3DMM-Fitting-Pytorch> python fit_single_img.py --img_path data/000002.jpg --res_folder results
------------ Options -------------
cache_folder: fitting_cache
exp_reg_w: 0.0008
first_nrf_iters: 500
first_rf_iters: 1000
gpu: 0
id_reg_w: 0.001
img_path: data/000002.jpg
lm_loss_w: 100
nframes_shape: 16
nrf_lr: 0.01
padding_ratio: 0.3
recon_model: bfm09
res_folder: results
rest_nrf_iters: 30
rest_rf_iters: 50
rf_lr: 0.01
rgb_loss_w: 1.6
rot_reg_w: 1
tar_size: 256
tex_reg_w: 1.7e-06
tex_w: 1
trans_reg_w: 1
-------------- End ----------------
loading models
loading images
image is loaded. width: 423, height: 594
C:\Users\weida\AppData\Local\Programs\Python\Python38\lib\site-packages\torch\nn\functional.py:780: UserWarning: Note that order of the arguments: ceil_mode and return_indices will changeto match the args list in nn.MaxPool2d in a future release.
  warnings.warn("Note that order of the arguments: ceil_mode and return_indices will change"
A face is detected. l: 44, t: 84, r: 356, b: 396
landmarks detected.
start rigid fitting
100%|█████████████████████████████████████████████████████████████████████████████| 1000/1000 [00:04<00:00, 239.56it/s]
done rigid fitting. lm_loss: 0.000491
start non-rigid fitting
  0%|                                                                                          | 0/500 [00:00<?, ?it/s]
Traceback (most recent call last):
  File "fit_single_img.py", line 158, in <module>
    fit(args)
  File "fit_single_img.py", line 78, in fit
    pred_dict = recon_model(recon_model.get_packed_tensors(), render=True)
  File "C:\Users\weida\AppData\Local\Programs\Python\Python38\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl
    return forward_call(*input, **kwargs)
  File "D:\AI\3DMM\3DMM-Fitting-Pytorch\core\BFM09Model.py", line 105, in forward
    rendered_img = self.renderer(mesh)
  File "C:\Users\weida\AppData\Local\Programs\Python\Python38\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl
    return forward_call(*input, **kwargs)
  File "d:\ai\3dmm\pytorch3d\pytorch3d\renderer\mesh\renderer.py", line 63, in forward
    fragments = self.rasterizer(meshes_world, **kwargs)
  File "C:\Users\weida\AppData\Local\Programs\Python\Python38\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl
    return forward_call(*input, **kwargs)
  File "d:\ai\3dmm\pytorch3d\pytorch3d\renderer\mesh\rasterizer.py", line 240, in forward
    pix_to_face, zbuf, bary_coords, dists = rasterize_meshes(
  File "d:\ai\3dmm\pytorch3d\pytorch3d\renderer\mesh\rasterize_meshes.py", line 234, in rasterize_meshes
    pix_to_face, zbuf, barycentric_coords, dists = _RasterizeFaceVerts.apply(
  File "d:\ai\3dmm\pytorch3d\pytorch3d\renderer\mesh\rasterize_meshes.py", line 308, in forward
    pix_to_face, zbuf, barycentric_coords, dists = _C.rasterize_meshes(
RuntimeError: Not compiled with GPU support

[bottler]

For some reason PyTorch3D built without cuda. You could try ensuring CUDA_HOME is set, creating a new environment. Then try to build pytorch3d again, setting the environment variable FORCE_CUDA to 1.

[wwdok]

I have successfully install pytorch3d with GPU support, let me summarize the process.

1. First i mannually set force_cuda to 1, but 'pip install -e .' still didn't execute def get_extensions(), then i doubt 'pip install -e .' did NOT execute setup.py at all, i insert some prints insides setup.py, and run 'pip install -e .', it still not print out any print statements, so maybe 'pip install -e .' really didn't execute setup.py.
2. I go to INSTALL.md， reread the guide, find there is anther install method : python setup.py install
   
   execxuting python setup.py install, this time the windows prompt print out those print statements i inserted, so it proves setup.py has been executed.
3. from the log, i found some warning and error, then i go to https://github.com/NVIDIA/cub/releases, download its source code, and set env variable CUB_HOME to its folder path, then i also use the x64 Native Tools Command Prompt for VS 2019 instead of common cmd prompt.