fast and aprox srgb to linear space

[lassade]

Hi,

I come up with a function that aproximates the gamma conversion is has a RMSE of 0.001064 which is better than the simple x^2 aproach here is the math: https://www.desmos.com/calculator/rbjqw6re8i

pub inline fn linearAprox(val: F32x4) F32x4 {
    var a = val * val;
    var b = a * @sqrt(val);
    b = f32x4s(0.567) * a + f32x4s(1 - 0.567) * b;
    return @shuffle(f32, b, val, [4]i32{ 0, 1, 2, ~@as(i32, 3) });
}

[michal-z]

Interesting. Please make a PR with some tests.

[mochalins]

I'm getting results that don't quite seem consistent with a RMSE of 0.001064, although that may be because I've happened to pick particularly large error test cases. PR/tests in #459, wasn't sure where to put it (would be an equivalent to function provided in stb_image), so just stuck it in zmath for now.

[lassade]

I would guess that the RMSE of 0.001064 is error smaller than the actual error because desmos uses a 64-bit float

You can change the 0.567 constant to see if the error improves using 32-bit float

this algorithm is quite silly, I don't think is worth using ...