Rust allocation

[sletz]

In the Rust backend:

pub struct mydsp {
    fSampleRate: i32,
    fConst0: F32,
    fHslider0: F32,
    fHslider1: F32,
    fRec0: [F32;3],
}

impl FaustDsp for mydsp {
    type T = F32;

    fn new() -> mydsp { 
        mydsp {
            fSampleRate: 0,
            fConst0: 0.0,
            fHslider0: 0.0,
            fHslider1: 0.0,
            fRec0: [0.0;3],
        }
    }
...

should be rewritten (for big arrays):

pub struct mydsp {
    fSampleRate: i32,
    fConst0: F32,
    fHslider0: F32,
    fHslider1: F32,
    fRec0: Box<[f32; 3]>,
}

impl FaustDsp for mydsp {
    type T = F32;

    fn new() -> mydsp { 
        mydsp {
            fSampleRate: 0,
            fConst0: 0.0,
            fHslider0: 0.0,
            fHslider1: 0.0,
            fRec0: vec![0.0; 3].into_boxed_slice(),
        }
    }
...

[bluenote10]

Hm, but then we enforce boxing implicitly, which could prevent usage in embedded scenarios and it could lead to worse performance due to reduced cache locality.

Wasn't this solved in https://github.com/grame-cncm/faust/pull/828 by allowing to opt-in to the default_boxed feature? This already works pretty well for large-sized DSPs.

[sletz]

OK, actually somewhat forgot that discussion ((-;... so nothing to do then.