eunomia 0.8.0

Atlas datatype law: the Single Source of Truth for numeric/scalar datatype vocabulary (scalars, complex, packed formats, conversions)
Documentation 
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
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173

//! `FloatElement` impls for the wrapper float types (native f64 via the
//! explicit F64 impl; reduced-precision types route through f32).

use crate::traits::FloatElement;
use crate::types::{Bf16, Bf4, Bf8, F16, F32, F4, F64, F8};

macro_rules! impl_float_element {
    ($t:ident, $from_f32:expr, $from_f64:expr, $to_f32:expr) => {
        impl FloatElement for $t {
            #[inline(always)]
            fn from_f32(val: f32) -> Self {
                $from_f32(val)
            }
            #[inline(always)]
            fn from_f64(val: f64) -> Self {
                $from_f64(val)
            }
            #[inline(always)]
            fn to_f32(self) -> f32 {
                $to_f32(self)
            }
        }
    };
}

impl_float_element!(F16, F16::from_f32, F16::from_f64, F16::to_f32);
impl_float_element!(F32, F32, |val| F32(val as f32), |x: F32| x.0);
// F64 wraps native `f64`, so it gets an explicit impl with native
// double-precision transcendentals — the macro's f32-routed default would
// widen-narrow and discard f64 precision. (F32 routes through f32 = native;
// F16/Bf16/F8/F4/Bf8/Bf4 have no hardware transcendentals, so the f32 default is
// their correct reduced-precision path.)
impl FloatElement for F64 {
    #[inline(always)]
    fn from_f32(val: f32) -> Self {
        F64(val as f64)
    }
    #[inline(always)]
    fn from_f64(val: f64) -> Self {
        F64(val)
    }
    #[inline(always)]
    fn to_f32(self) -> f32 {
        self.0 as f32
    }
    #[inline]
    fn exp(self) -> Self {
        F64(libm::exp(self.0))
    }
    #[inline]
    fn ln(self) -> Self {
        F64(libm::log(self.0))
    }
    #[inline]
    fn sin(self) -> Self {
        F64(libm::sin(self.0))
    }
    #[inline]
    fn cos(self) -> Self {
        F64(libm::cos(self.0))
    }
    #[inline]
    fn acos(self) -> Self {
        F64(libm::acos(self.0))
    }
    #[inline]
    fn tan(self) -> Self {
        F64(libm::tan(self.0))
    }
    #[inline]
    fn sinh(self) -> Self {
        F64(libm::sinh(self.0))
    }
    #[inline]
    fn cosh(self) -> Self {
        F64(libm::cosh(self.0))
    }
    #[inline]
    fn tanh(self) -> Self {
        F64(libm::tanh(self.0))
    }
    #[inline]
    fn atan2(self, other: Self) -> Self {
        F64(libm::atan2(self.0, other.0))
    }
    #[inline]
    fn powf(self, n: Self) -> Self {
        F64(libm::pow(self.0, n.0))
    }
    #[inline]
    fn recip(self) -> Self {
        F64(1.0 / self.0)
    }
    #[inline]
    fn floor(self) -> Self {
        F64(libm::floor(self.0))
    }
    #[inline]
    fn ceil(self) -> Self {
        F64(libm::ceil(self.0))
    }
    #[inline]
    fn round(self) -> Self {
        F64(libm::round(self.0))
    }
    #[inline]
    fn trunc(self) -> Self {
        F64(libm::trunc(self.0))
    }
    #[inline]
    fn signum(self) -> Self {
        if self.0.is_nan() {
            self
        } else {
            F64(libm::copysign(1.0, self.0))
        }
    }
}
impl_float_element!(Bf16, Bf16::from_f32, Bf16::from_f64, Bf16::to_f32);
impl_float_element!(
    Bf8,
    Bf8::from_f32,
    |val| Bf8::from_f32(val as f32),
    |x: Bf8| x.to_f32()
);
impl_float_element!(
    Bf4,
    Bf4::from_f32,
    |val| Bf4::from_f32(val as f32),
    |x: Bf4| x.to_f32()
);
impl_float_element!(F8, F8::from_f32, |val| F8::from_f32(val as f32), |x: F8| x
    .to_f32());
impl_float_element!(F4, F4::from_f32, |val| F4::from_f32(val as f32), |x: F4| x
    .to_f32());

#[cfg(test)]
mod tests {
    use crate::traits::FloatElement;
    use crate::types::F64;

    #[test]
    fn f64_wrapper_transcendentals_are_native_precision() {
        // Native f64 agrees with std f64 to ~machine epsilon; the f32-routed
        // default would only be accurate to ~1e-7, failing these bounds.
        assert!((F64(1.0).exp().0 - core::f64::consts::E).abs() < 1e-15);
        assert!((F64(0.1).ln().0 - 0.1_f64.ln()).abs() < 1e-15);
        assert!((F64(0.7).sin().0 - 0.7_f64.sin()).abs() < 1e-15);
        assert!((F64(0.25).acos().0 - 0.25_f64.acos()).abs() < 1e-15);
        assert!((F64(2.0).powf(F64(10.0)).0 - 1024.0).abs() < 1e-12);
    }

    #[test]
    fn rounding_and_powi_surface() {
        // UFCS calls the FloatElement impl explicitly (concrete f64 would
        // otherwise resolve to std's inherent methods, not eunomia's).
        assert_eq!(FloatElement::floor(2.7_f64), 2.0);
        assert_eq!(FloatElement::ceil(2.2_f64), 3.0);
        assert_eq!(FloatElement::round(2.5_f64), 3.0);
        assert_eq!(FloatElement::trunc(-2.7_f64), -2.0);
        assert!((FloatElement::acos(0.25_f64) - 0.25_f64.acos()).abs() < 1e-15);
        // signum: ±1 with the sign (num_traits / std semantics), NaN→NaN.
        assert_eq!(FloatElement::signum(-3.5_f64), -1.0);
        assert_eq!(FloatElement::signum(0.0_f64), 1.0);
        // powi: exact integer power, correct for negative base + exponent.
        assert_eq!(FloatElement::powi(2.0_f64, 10), 1024.0);
        assert_eq!(FloatElement::powi(-2.0_f64, 3), -8.0);
        assert!((FloatElement::powi(2.0_f64, -2) - 0.25).abs() < 1e-15);
        // F64 wrapper (native impl).
        assert_eq!(FloatElement::floor(F64(2.7)).0, 2.0);
        assert_eq!(FloatElement::powi(F64(2.0), 10).0, 1024.0);
    }
}