Resisting the urge to write an F16x8 SIMD type and see how much I can implement using AVX2 (no extensions)
That just converts though, right? I'm talking about implementing all the arithmetic ops natively on F16. Not likely to work well, but would be fun to implement :)
-
-
You can implement all of the arithmetic by lifting to f32, doing the op, and converting back.
-
This gives a correctly-rounded result for every operation except for FMA (converting to f64 suffices there, though you need to be careful about rounding back to f16).
End of conversation
New conversation -
-
-
Godspeed
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.