setup.py on Fedora 30

[djhenry]

Describe the bug When running python setup.py on Fedora 30, cudnn_checkfiles_linux() is returning cudnn_checkfiles with path /include64 rather than /usr/include causing setup to incorrectly report "ERROR cuDNN not found. See https://github.com/deepfakes/faceswap/blob/master/INSTALL.md#cudnn for instructions".

I was able to work around by creating a symlink to that path ln -s /usr/include /include64

I notice this cudnn_checkfiles_linux() is also duplicated in lib/sysinfo.py

To Reproduce Steps to reproduce the behavior: Follow install steps found at https://github.com/deepfakes/faceswap/blob/master/INSTALL.md#user-content-general-install-guide (Not using Docker, using conda environment, Negativo repositories for nvidia and cuda driver, nvidia repositories for nvcc and libcudnn)

Expected behavior Setup success with cuda enabled

Desktop (please complete the following information): encoding: UTF-8 git_branch: master git_commits: 967f97f Minor extract fixups gpu_cuda: 10.1 gpu_cudnn: 7.6.3 gpu_devices: GPU_0: GeForce RTX 2080 gpu_devices_active: GPU_0 gpu_driver: 430.40 gpu_vram: GPU_0: 7979MB os_machine: x86_64 os_platform: Linux-5.2.9-200.fc30.x86_64-x86_64-with-fedora-30-Thirty os_release: 5.2.9-200.fc30.x86_64 py_command: faceswap.py gui py_conda_version: N/A py_implementation: CPython py_version: 3.7.4 py_virtual_env: False sys_cores: 6 sys_processor: x86_64 sys_ram: Total: 15972MB, Available: 6881MB, Used: 8534MB, Free: 294MB

[torzdf]

Feel free to raise a PR for checking of cuDNN on Fedora.

Unfortunately different OSes have it in different places, so it is hard to capture them all

[djhenry]

I'm relatively new to python. I'll see what I can do.

[torzdf]

I'm closing this issue as the recommended install method for Linux is to use the install script:

https://github.com/deepfakes/faceswap/releases/tag/v1.0.0