memory: add cuda 13 support

[maxime-peim]

In the latest cuda driver API, cuCtxCreate takes another parameter CUctxCreateParams.

[vincentkfu]

This will fail if the build environment has a CUDA version before 13.0. We should have a test in the configure script that detects whether this additional parameter is required and uses the new function signature only if it is required.

Here is an example of a similar change: f95cf49

Also please add to the bottom of the commit message:

Fixes: 1988

[sitsofe]

@maxime-peim Could you also change the commit's title to something like

memory: add cuda 13 support 

[maxime-peim]

Sure, let me add these

[axboe]

I don't see how this works, as nothing looks at this exit value. In other words, this will end up being true if it compiles, which presumably it'll do for both version 13 or below, not really checking what you'd want.

Why isn't this just doing

cuCtxCreate(NULL, NULL, 0, NULL);

and check if that compiles (which is what compile_prog does)? We already know we have cuda when this test is compiled, hence we just need to check if a 4-parameter cuCtxCreate compiles or not. If it does, hopefully this is cuda >= 13, if it doesn't then it should be < 13.

[maxime-peim]

Totally, sorry, I don't know why I thought that if compilation passes then we execute the resulting binary to check the results.
I'll update the PR

[axboe]

That looks better!