how to define fp16 using macro

Accelerated Computing CUDA CUDA Programming and Performance
#1

Hello,

I want to use PI as half precision macro. I can do it in float as below.

#define PI 3.14159265359f

I can use the form below.

#define PI __float2half(3.14159265359f)

Is there a more "native" implementation in half precision, where no extra conversion is needed?

Thanks in advance!