1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134

use super::*;

#[const_trait]
pub trait ApproxInvSqrt
{
    /// Calculates an approximation of 1/sqrt(x).
    /// 
    /// The result gets iterated through the Newton-Raphson method with a given amount of iterations.
    /// 
    /// At 5 or less iterations, it is generally faster than x.sqrt().recip() in release build.
    /// 
    /// 1 iteration was good enough for quake.
    /// 
    /// # Example
    /// 
    /// ```rust
    /// #![feature(const_trait_impl)]
    /// 
    /// use float_approx_math::ApproxInvSqrt;
    /// 
    /// const X: f32 = 2.0;
    /// let y: f32 = X.approx_inv_sqrt::<4>(); // Three iterations
    ///
    /// assert_eq!(y, X.sqrt().recip());
    /// ```
    fn approx_inv_sqrt<const NEWTON: usize>(self) -> Self;
    
    fn approx_inv_sqrt_unchecked<const NEWTON: usize>(self) -> Self;
}

macro_rules! impl_approx_inv_sqrt {
    ($float:ty: $bits:ty; $consts:tt) => {
        impl const ApproxInvSqrt for $float
        {
            fn approx_inv_sqrt<const NEWTON: usize>(self) -> Self
            {
                if self > 0.0
                {
                    self.approx_inv_sqrt_unchecked::<NEWTON>()
                }
                else
                {
                    <$float>::NAN
                }
            }
            fn approx_inv_sqrt_unchecked<const NEWTON: usize>(self) -> Self
            {
                const L: $bits = 1 << (<$float>::MANTISSA_DIGITS as $bits - 1);
                const SIGMA: f64 = 0.0450466;
                const MAGIC_NUMBER: $bits = (1.5*L as f64*($consts::EXP_BIAS as f64 - SIGMA) + 0.5) as $bits;

                let mut y = <$float>::from_bits(MAGIC_NUMBER - (<$float>::to_bits(self) >> 1));

                let x2 = self*0.5;
                let mut i = 0;
                while i < NEWTON
                {
                    y *= 1.5 - x2*y*y;
                    i += 1;
                }
                y
            }
        }
    };
}

#[cfg(test)]
mod test
{
    use ::test::Bencher;

    use super::*;
    use crate::tests as t;

    #[test]
    fn verify_magic_number_f32()
    {
        const L: u32 = 1 << (<f32>::MANTISSA_DIGITS - 1);
        const SIGMA: f64 = 0.0450466;
        const MAGIC_NUMBER: u32 = (1.5*L as f64*(f32::EXP_BIAS as f64 - SIGMA) + 0.5) as u32;
    
        assert_eq!(MAGIC_NUMBER, 0x5f3759df);
    }
    
    #[test]
    fn inv_sqrt()
    {
        const RANGE: f32 = 10.0;
        
        t::plot_approx("inv_sqrt_0", 0.1..RANGE, |x| x.sqrt().recip(), ApproxInvSqrt::approx_inv_sqrt::<0>);
        t::plot_approx("inv_sqrt_1", 0.1..RANGE, |x| x.sqrt().recip(), ApproxInvSqrt::approx_inv_sqrt::<1>);
        t::plot_approx("inv_sqrt_2", 0.1..RANGE, |x| x.sqrt().recip(), ApproxInvSqrt::approx_inv_sqrt::<2>);
        t::plot_approx("inv_sqrt_3", 0.1..RANGE, |x| x.sqrt().recip(), ApproxInvSqrt::approx_inv_sqrt::<3>);
    }
    
    #[bench]
    fn inv_sqrt_benchmark(_: &mut Bencher)
    {
        type F = f64;

        const N: usize = 1000;
        const S: usize = 32;

        t::plot_benchmark::<_, _, N, _>(
            "inv_sqrt",
            [
                &|x: F| x.sqrt().recip(),
                &ApproxInvSqrt::approx_inv_sqrt::<0>,
                &ApproxInvSqrt::approx_inv_sqrt::<1>,
                &ApproxInvSqrt::approx_inv_sqrt::<2>,
                &ApproxInvSqrt::approx_inv_sqrt::<3>,
                &ApproxInvSqrt::approx_inv_sqrt::<4>
            ],
            0.1..256.0,
            S
        );

        {
            const N: usize = 1000000;
            let x: Vec<F> = (0..N).map(|i| (i + 1) as F).collect();
    
            println!("sqrt dt = {:?}", t::benchmark(&x, &|x: F| x.sqrt().recip()));
            
            println!("approx_inv_sqrt::<0> dt = {:?}", t::benchmark(&x, &|x| x.approx_inv_sqrt::<0>()));
            println!("approx_inv_sqrt::<1> dt = {:?}", t::benchmark(&x, &|x| x.approx_inv_sqrt::<1>()));
            println!("approx_inv_sqrt::<2> dt = {:?}", t::benchmark(&x, &|x| x.approx_inv_sqrt::<2>()));
            println!("approx_inv_sqrt::<3> dt = {:?}", t::benchmark(&x, &|x| x.approx_inv_sqrt::<3>()));
            println!("approx_inv_sqrt::<4> dt = {:?}", t::benchmark(&x, &|x| x.approx_inv_sqrt::<4>()));
        }
    }
}

impl_approx_inv_sqrt!(f32: u32; f32);
impl_approx_inv_sqrt!(f64: u64; f64);