cudaCheckError() failed: initialization error

[yingxin666]

when I run "output = _prroi_pooling.prroi_pooling_forward_cuda(features, rois, *params)" ,I get the debug "cudaCheckError() failed: initialization error".

[vacancy]

        pool = PrRoIPool2D(7, 7, spatial_scale=0.5)
        features = torch.rand((4, 16, 24, 32)).cuda()
        rois = torch.tensor([
            [0, 0, 0, 14, 14],
            [1, 14, 14, 28, 28],
        ]).float().cuda()

        out = pool(features, rois)

Can you check if this minimal script run on your machine? The error you encountered seems like a pytorch-level error. Can you check if built-in methods, for example, output = torch.nn.functional.avg_pool2d(features, 7) work?

[yingxin666]

"out = pool(features, rois)" can not work,the error is the same.but "output = torch.nn.functional.avg_pool2d(features, 7)" can work well.

[yingxin666]

the full error message is:
Using /home/zyd/.cache/torch_extensions as PyTorch extensions root...
Detected CUDA files, patching ldflags
Emitting ninja build file /home/zyd/.cache/torch_extensions/_prroi_pooling/build.ninja...
Building extension module _prroi_pooling...
Allowing ninja to set a default number of workers... (overridable by setting the environment variable MAX_JOBS=N)
ninja: no work to do.
Loading extension module _prroi_pooling...
cudaCheckError() failed : initialization error

[vacancy]

I see. I think this is probably a issue with your cuda compiler setup etc. Is it possible to run the minimal code example here: https://pytorch.org/tutorials/advanced/cpp_extension.html (The CUDA extension part) ?