I was just wondering, why didn’t Nvidia implement operator overloads for FP16 and __half2 data types? +, -, comparsions etc…
Because I’m writing templatised kernels which can operate on floats and on __half data types and I’m getting tired of doing “if constexpr” (c++ 17 feature which gives me warnings when compiling kernels) just to do a simple __half addition using intrinsics when I can just write “a + b”.
Can I even implement my own operator overloads?
Kind regard,
Garko