He mentioned "performance cost". 8 byte floats are twice as big, half as
SIMD-vectorizable, and often more precision than you need (especially in video
games which he also mentioned). I know a couple people who advocate 16bit/2B
floats (even outside of neural nets!).
Well it's mostly a matter of avoiding the default type of float on your
platform (which is `float64` on a 64-bit machine). The same thing happens if
you for some reason decides you don't want to just use `int` and instead try to
make everything `int16` for example.
Why do you want to use `float
A bonus one:
var a = 1
var b = 2
var c: float32 = a / b
# compiles but you better write:
var d = a.float32 / b.float32
Run
Hello,
A thing that confuses me when writing game code is using float32 types
correctly (so it does compile) while avoiding too many explicit 'to float32'
conversions. Another important point is to prevent implicit conversions between
float32 and float64, when doing arithmetic because as I unde
