Why fixed point instead of float?

[obsoleszenz]

Is there any reason to use fixed point over float? I thought it was an optimization of the past, but maybe I'm misinformed?

[Holzhaus]

Sorry, I'm lacking context here. Where is fixed point arithmetic used?

[obsoleszenz]

Maybe i'm just assuming it, but isn't this fixed point arithmetic?

https://github.com/Holzhaus/vinylla/blob/main/src/util.rs#L68

Also here the sample is a i32 ? I would have expected an f32 ?

vinylla/src/timecode.rs

Line 72 in 659fe1e

self.ewma.process(sample);

[Holzhaus]

Maybe i'm just assuming it, but isn't this fixed point arithmetic?

https://github.com/Holzhaus/vinylla/blob/main/src/util.rs#L68

No, that's just integer math. Not using fixed point arithmeric anywhere IIRC.

Also here the sample is a i32 ? I would have expected an f32 ?

vinylla/src/timecode.rs

Line 72 in 659fe1e

self.ewma.process(sample);

Originally the samples are an i16, because 16 bit audio should suffice for our purposes. The Serato timecode audio CDs are also 16 bit anyway.

https://github.com/Holzhaus/vinylla/blob/main/src%2Ftimecode.rs#L129-L130

Using integers here because it's faster and I don't see an advantage in using floating point math for this use case.