simdeez 3.0.1

SIMD library to abstract over different instruction sets and widths
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

use crate::math::{f32, f64, map, scalar};
use crate::{Simd, SimdFloat32, SimdFloat64};

// DECISION(2026-03-23): KEEP_SCALAR_REFERENCE
// Function(s): f32 ln_u35 / exp_u35
// Why scalar:
// - the final combined-wave recheck still keeps runtime-selected ln_u35 and exp_u35 below native scalar
//   on this host (`ln`: about 2.72 ms vs 2.46 ms, `exp`: about 2.34 ms vs 2.11 ms)
// - these contracts are stricter than the relaxed portable f32 log2_u35 / exp2_u35 pieces they would
//   naturally compose from, so there is no cheap honest rescue today
// Revisit when:
// - a better shared f32 log/exp kernel exists

pub trait SimdMathF32Core: SimdFloat32 {
    #[inline(always)]
    fn log2_u35(self) -> Self
    where
        Self::Engine: Simd<Vf32 = Self>,
    {
        f32::log2_u35(self)
    }

    #[inline(always)]
    fn exp2_u35(self) -> Self
    where
        Self::Engine: Simd<Vf32 = Self>,
    {
        f32::exp2_u35(self)
    }

    #[inline(always)]
    fn ln_u35(self) -> Self {
        map::unary_f32(self, scalar::ln_u35_f32)
    }

    #[inline(always)]
    fn exp_u35(self) -> Self {
        map::unary_f32(self, scalar::exp_u35_f32)
    }

    #[inline(always)]
    fn sin_u35(self) -> Self
    where
        Self::Engine: Simd<Vf32 = Self>,
    {
        f32::sin_u35(self)
    }

    #[inline(always)]
    fn cos_u35(self) -> Self
    where
        Self::Engine: Simd<Vf32 = Self>,
    {
        f32::cos_u35(self)
    }

    #[inline(always)]
    fn tan_u35(self) -> Self
    where
        Self::Engine: Simd<Vf32 = Self>,
    {
        f32::tan_u35(self)
    }
}

impl<T: SimdFloat32> SimdMathF32Core for T {}

pub trait SimdMathF64Core: SimdFloat64 {
    #[inline(always)]
    fn log2_u35(self) -> Self
    where
        Self::Engine: Simd<Vf64 = Self>,
    {
        f64::log2_u35(self)
    }

    #[inline(always)]
    fn exp2_u35(self) -> Self
    where
        Self::Engine: Simd<Vf64 = Self>,
    {
        f64::exp2_u35(self)
    }

    #[inline(always)]
    fn ln_u35(self) -> Self
    where
        Self::Engine: Simd<Vf64 = Self>,
    {
        f64::ln_u35(self)
    }

    #[inline(always)]
    fn exp_u35(self) -> Self
    where
        Self::Engine: Simd<Vf64 = Self>,
    {
        f64::exp_u35(self)
    }

    #[inline(always)]
    fn sin_u35(self) -> Self
    where
        Self::Engine: Simd<Vf64 = Self>,
    {
        f64::sin_u35(self)
    }

    #[inline(always)]
    fn cos_u35(self) -> Self
    where
        Self::Engine: Simd<Vf64 = Self>,
    {
        f64::cos_u35(self)
    }

    #[inline(always)]
    fn tan_u35(self) -> Self
    where
        Self::Engine: Simd<Vf64 = Self>,
    {
        f64::tan_u35(self)
    }
}

impl<T: SimdFloat64> SimdMathF64Core for T {}