[PyTorch][CUDNN][Windows] Version 1.22.1 fails to compile with NV_CUDNN_FRONTEND_USE_DYNAMIC_LOADING

[Anerudhan]

Attempted of doing NV_CUDNN_FRONTEND_USE_DYNAMIC_LOADING ends up in "Unable to find symbol cudnnBackendCreateDescriptor".
Upstream PyTorch pull request: pytorch/pytorch#179820
Full error message prints:

2026-04-13T11:50:51.6448543Z C:\actions-runner\_work\pytorch\pytorch\.ci\pytorch\windows>echo Checking that basic CNN works
2026-04-13T11:50:51.6450693Z Checking that basic CNN works
2026-04-13T11:50:51.6457093Z
2026-04-13T11:50:51.6458945Z C:\actions-runner\_work\pytorch\pytorch\.ci\pytorch\windows>"python3.14t" C:\actions-runner\_work\pytorch\pytorch\.ci\pytorch\test_example_code\cnn_smoke.py
2026-04-13T11:50:53.7188412Z C:\actions-runner\_work\pytorch\pytorch\.ci\pytorch\windows\Python\Lib\site-packages\torch\_subclasses\functional_tensor.py:362: UserWarning: Failed to initialize NumPy: No module named 'numpy' (Triggered internally at C:\actions-runner\_work\pytorch\pytorch\torch\csrc\utils\tensor_numpy.cpp:84.)
2026-04-13T11:50:53.7196091Z   cpu = _conversion_method_template(device=torch.device("cpu"))
2026-04-13T11:50:54.9957972Z Traceback (most recent call last):
2026-04-13T11:50:54.9978543Z   File "C:\actions-runner\_work\pytorch\pytorch\.ci\pytorch\test_example_code\cnn_smoke.py", line 28, in <module>
2026-04-13T11:50:54.9979730Z     outputs = net(net_inputs)
2026-04-13T11:50:54.9981312Z   File "C:\actions-runner\_work\pytorch\pytorch\.ci\pytorch\windows\Python\Lib\site-packages\torch\nn\modules\module.py", line 1778, in _wrapped_call_impl
2026-04-13T11:50:54.9982793Z     return self._call_impl(*args, **kwargs)
2026-04-13T11:50:54.9983455Z            ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
2026-04-13T11:50:54.9984773Z   File "C:\actions-runner\_work\pytorch\pytorch\.ci\pytorch\windows\Python\Lib\site-packages\torch\nn\modules\module.py", line 1789, in _call_impl
2026-04-13T11:50:54.9986192Z     return forward_call(*args, **kwargs)
2026-04-13T11:50:54.9987311Z   File "C:\actions-runner\_work\pytorch\pytorch\.ci\pytorch\test_example_code\cnn_smoke.py", line 19, in forward
2026-04-13T11:50:54.9988559Z     output = self.pool(F.relu(self.conv(inputs)))
2026-04-13T11:50:54.9989276Z                               ~~~~~~~~~^^^^^^^^
2026-04-13T11:50:54.9990678Z   File "C:\actions-runner\_work\pytorch\pytorch\.ci\pytorch\windows\Python\Lib\site-packages\torch\nn\modules\module.py", line 1778, in _wrapped_call_impl
2026-04-13T11:50:54.9992119Z     return self._call_impl(*args, **kwargs)
2026-04-13T11:50:54.9992717Z            ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
2026-04-13T11:50:54.9994029Z   File "C:\actions-runner\_work\pytorch\pytorch\.ci\pytorch\windows\Python\Lib\site-packages\torch\nn\modules\module.py", line 1789, in _call_impl
2026-04-13T11:50:55.0000302Z   File "C:\actions-runner\_work\pytorch\pytorch\.ci\pytorch\windows\Python\Lib\site-packages\torch\nn\modules\conv.py", line 565, in forward
2026-04-13T11:50:55.0002221Z     return self._conv_forward(input, self.weight, self.bias)
2026-04-13T11:50:55.0003960Z            ~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2026-04-13T11:50:55.0005678Z   File "C:\actions-runner\_work\pytorch\pytorch\.ci\pytorch\windows\Python\Lib\site-packages\torch\nn\modules\conv.py", line 560, in _conv_forward
2026-04-13T11:50:55.0007408Z     return F.conv2d(
2026-04-13T11:50:55.0008029Z            ~~~~~~~~^
2026-04-13T11:50:55.0008971Z         input, weight, bias, self.stride, self.padding, self.dilation, self.groups
2026-04-13T11:50:55.0010278Z         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2026-04-13T11:50:55.0011107Z     )
2026-04-13T11:50:55.0011628Z     ^
2026-04-13T11:50:55.0012387Z RuntimeError: Unable to find symbol cudnnBackendCreateDescriptor