Is there a native_divide for half_n?

[fangqq]

it is not clear from the online documentation

https://www.khronos.org/registry/Ope...ml/divide.html

but when I call it, JIT compiler failed to compile native_divide(half4, half4) (on AMD amdgpu-pro 17.70, vega64 GPU). is halfn supported by the native_divide function?

I've already have the

#pragma OPENCL EXTENSION cl_khr_fp16 : enable

in the code.