Arch Linux

Thanks for the report. Is there a chance you can make a patch? If not, I'll take a look but it'll likely take some time.

Hi,

It would be awesome to include this. I have uploaded the .pc files I personally use for CUDA 10.1, which work *for me*, just in case anybody else needs them.

@Svenstaro, it would be awesome to have this included in the ABS. I can spend a little bit of time making a patch and testing it if needed.

> They are required to build gst-plugins-bad's nvenc plugin on archlinux.

Where do these pc files usually come from, if they are missing in our package?

Uploading some personal files you happen to use, seems like a recipe for disaster in terms of reliability and consistency with upstream expectations and the wider cross-distro ecosystem. Is there one somewhere in the official upstream source code?

I'd be happy to include them but I'd like to rather extract them directly from the package somehow. I know they are there, just perhaps tedious to get to.

> Where do these pc files usually come from, if they are missing in our package?

They are created by the cuda-installer, AFAIK.

> Uploading some personal files you happen to use, seems like a recipe for disaster in terms of reliability and consistency with upstream expectations and the wider cross-distro ecosystem. Is there one somewhere in the official upstream source code?

There is no 'upstream source code'. This package is built from closed-source binary distributed files, that's the point. This files are generated in a binary-only distributed installer, so we cannot infer what they have without running the installer. So either the installer is actually ran in a fakeroot environment (or whatever), and then the files are copied, or they are generated by the PKGBUILD.

Whatever approach is preferred is fine for me, I just uploaded this because currently any software that uses pkg-config to figure out the CUDA RTA/Driver location is broken in Arch Linux.

> I'd be happy to include them but I'd like to rather extract them directly from the package somehow. I know they are there, just perhaps tedious to get to.

Yes, they are there, but it requires running the cuda-installer instead of just extracting the package from the .run.

If I install cuda from nvidia's official repo on centos, it provides several *.pc file at /usr/lib64/pkgconfig:
cublas-10.pc
cuda-10.2.pc
cudart-10.2.pc
cufft-10.2.pc
cufftw-10.2.pc
cuinj64-10.2.pc
curand-10.2.pc
cusolver-10.2.pc
cusparse-10.2.pc

Which is own by package:
libcublas-devel-10.2.2.89-1.x86_64
cuda-driver-dev-10-2-10.2.89-1.x86_64
cuda-cudart-dev-10-2-10.2.89-1.x86_64
cuda-cufft-dev-10-2-10.2.89-1.x86_64
cuda-cufft-dev-10-2-10.2.89-1.x86_64
cuda-nvprof-10-2-10.2.89-1.x86_64
cuda-curand-dev-10-2-10.2.89-1.x86_64
cuda-cusolver-dev-10-2-10.2.89-1.x86_64
cuda-cusparse-dev-10-2-10.2.89-1.x86_64

These packages can be downloaded at nvidia's official cuda repo: "https://developer.download.nvidia.com/compute/cuda/repos/rhel8/x86_64/"

I whipped up a bunch of pkgconfig files based on the official ones provided by NVIDIA with a few minor Arch-specific modifications. I'll consider the bug closed. Request a re-open if I missed something.