Use std::fma ?

[nestordemeure]

Hello,

For floating point types, the standard implementation of std::fma guarantees that it :

Computes (x*y) + z as if to infinite precision and rounded only once to fit the result type.

In practice it uses an FMA instruction when available or an algorithm similar to your handmade implementation.

Wouldn't it make sense to use it in libeft (avoiding the dependency on intrinsics) ?

Cordially,
Nestor Demeure