CPack rpm and failed dependencies when install

Usage
1

In my project, I use below code to import cuda library

  find_package(CUDAToolkit)
  target_link_libraries(${target} PRIVATE CUDA::cudart)
  target_link_libraries(${target} PRIVATE CUDA::cufft)
  target_link_libraries(${target} PRIVATE CUDA::cublas)

It works well, and make rpm using cpack.

When I install the rpm package, it failed:

error: Failed dependencies:
	libcublas.so.11()(64bit) is needed by hello-1.0.0-1.x86_64
	libcublasLt.so.11()(64bit) is needed by hello-1.0.0-1.x86_64
	libcudart.so.11.0()(64bit) is needed by hello-1.0.0-1.x86_64
	libcufft.so.10()(64bit) is needed by hello-1.0.0-1.x86_64

I have installed cuda in /usr/local/cuda, and set path in /etc/ld.so.conf.d/cuda-11-4.conf.

Any idea? Thank you very mach!

2

Your CUDA installation is not tracked by RPM, so it won’t “see” them. Something needs to tell RPM that these CUDA libraries exist and are associated with some package. I suspect you’ll need to use an RPM installer for CUDA rather than a shell script.

Cc: @robert.maynard

3

I am not familar with the requirements of RPM, but yes CUDA does offer RPM packages.

If you go with a solution where you package CUDA dependencies, you need to read the CUDA EULA agreement as it has specific redist rules.