Nvidia Packaging

[Graham Inggs]

On 23/09/2014 10:41, Graham Inggs wrote:
> I've been unravelling some of the dependencies of nvvp, and should 
> have a patch for #761363 ready soon.
> There are also some changes to the packaging I'd like to propose, so I 
> think it's best if I attach patches to the list so they can be discussed.
> We should be able to upload 6.0 to unstable soon.

I see Andreas pushed a fix for #761363 and moved the dependency on 
libcuinj* to nvidia-profiler.

Remaining changes I want to suggest:

1. ship libcuinj{32,64}.so in nvidia-cuda-dev instead of libcuinj*
     move breaks/replaces on libjcuinj* to nvidia-cuda-dev
Reason: nvvp/nvprof haven't used libcuinj{32,64}.so since CUDA 4.1

2. upgrade recommends on ${package:libcuda1} to depends in nvidia-cuda-dev
Reason: if an application builds using the CUDA driver API and needs to 
be linked to libcuda.so, it should be available

3. downgrade depends on nvidia-libopencl-dev | opencl-dev (or remove 
completely) in nvidia-cuda-toolkit
Reason: is there any reason we need the opencl-dev stuff?

Let me know if I can work on any of these and whether they are wanted in 
the 5.5 branch as well.