Search code examples
c++cudathrust

Can thrust deal with half precision floats (float16 / IEEE 754-2008)?

I'm looking at an application that does not require 32bit of precision, 12-16bit will be enough.

Is thrust capable of dealing with float16's (i.e. packing/unpacking two 16bit precision floats into a 32bit word)?

Should I use fixed-point arithmetic?

Solution

  • CUDA hardware does not include native support for half-precision arithmetic, only conversion to and from float.

    Since C does not have a built-in half type, the conversion intrinsics use unsigned short:

    unsigned short __float2half_rn( float );
float __half2float( unsigned short );