Function f16x2_fma

pub unsafe fn f16x2_fma(a: f16x2, b: f16x2, c: f16x2) -> f16x2

Fused multiply-add, round to nearest even

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

Corresponds to the CUDA C intrinsics: