Function f16x2_max_nan

pub unsafe fn f16x2_max_nan(a: f16x2, b: f16x2) -> f16x2

Find the maximum of two values, NaNs pass through.

https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max

Corresponds to the CUDA C intrinsic __hmax2_nan