[FEA]: Provide generic and safe C++ interfaces for warp shuffle

[fbusato]

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

libcu++

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

CUDA provides warp shuffle intrinsics that support a limited set of types. Secondly, they there are not check to validate the inputs
see https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#warp-shuffle-functions

Describe the solution you'd like

Provide:

• cuda::shfl(T var, int srcLane, unsigned mask = 0xFFFFFFFF, int width=warpSize)
• cuda::shfl_up(T var, int delta, unsigned mask = 0xFFFFFFFF, int width=warpSize)
• cuda::shfl_down(T var, int delta, unsigned mask = 0xFFFFFFFF, int width=warpSize)
• cuda::shfl_xor(T var, int delta, unsigned mask = 0xFFFFFFFF, int width=warpSize)

that work with any trivially copyable types and validate the inputs, e.g. width must be a power of two, in debug mode

Describe alternatives you've considered

I suggest putting mask at the end because most of the time warp shuffles are used with all warp threads

Additional context

No response