When using float and double floating-point types for standard math library operations, there are detailed ULP error standards documented (as referenced in the CUDA C Programming Guide).
However, when I use the half floating-point type for standard math library operations, I lose access to the ULP error standards, as I couldn’t find related documentation on the official website.
Could you please advise how I can establish ULP error standards for the half type in standard math library operations? If I have missed any existing standards, kindly provide the relevant documentation.
Finally, could you confirm if the ULP error standard for the aforementioned half math functions and other half type math operations is actually 0?