[ENH]: Provide a helper for NVML "struct version" handling

[mdboom]

Is this a duplicate?

• I confirmed there appear to be no duplicate issues for this request and that I agree to the Code of Conduct

Area

cuda.bindings

Is your feature request related to a problem? Please describe.

NVML has a common pattern for specifying the version of struct passed to a function:

(<nvmlType_v2_t*>my_struct).version = sizeof(nvmlType_v2_t) | (2 << 24)

In the NVML header, this is handled by the NVML_STRUCT_VERSION macro:

#define NVML_STRUCT_VERSION(data, ver) (unsigned int)(sizeof(nvml ## data ## _v ## ver ## _t) | \
                                                      (ver << 24U))

We can't use this macro directly from cuda-bindings because:

1. This is a macro and we don't include the nvml.h header directly
2. Passing a struct name to a function and calling sizeof on it won't work in Cython

Describe the solution you'd like

There are still some ways this could be made slightly better and more obvious as to what is going on, however. Probably by adding an inline function that does:

nvmlStructVersion(unsigned int sizeof_struct, unsigned int ver)

Called like so:

version = nvmlStructVersion(sizeof(nvmlStruct_v2_t), 2))

Describe alternatives you've considered

No response

Additional context

No response