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
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144

//! f32 SIMD math kernel layering:
//! - `portable`: backend-agnostic reduction/polynomial kernels + scalar lane patching.
//! - `x86_avx2`: the only currently justified backend override (`log2_u35`).
//! - this module: dispatch glue selecting overrides without changing the public API.

mod hyperbolic;
mod portable;

#[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
mod x86_avx2;

use crate::{Simd, SimdFloat32};

// DECISION(2026-03-23): KEEP_SIMD_OVERRIDE
// Function(s): f32 log2_u35 dispatch
// Why kept:
// - local benches show the AVX2 path materially ahead of native scalar and forced scalar
// - the portable fallback also stays worthwhile on non-AVX2 backends
// Revisit when:
// - the portable log2 kernel or the AVX2 override changes materially

#[inline(always)]
pub(crate) fn log2_u35<V>(input: V) -> V
where
    V: SimdFloat32,
    V::Engine: Simd<Vf32 = V>,
{
    #[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
    {
        if is_avx2_engine::<V::Engine>() {
            return unsafe { x86_avx2::log2_u35(input) };
        }
    }

    portable::log2_u35(input)
}

#[inline(always)]
pub(crate) fn exp2_u35<V>(input: V) -> V
where
    V: SimdFloat32,
    V::Engine: Simd<Vf32 = V>,
{
    portable::exp2_u35(input)
}

#[inline(always)]
pub(crate) fn sin_u35<V>(input: V) -> V
where
    V: SimdFloat32,
    V::Engine: Simd<Vf32 = V>,
{
    portable::sin_u35(input)
}

#[inline(always)]
pub(crate) fn cos_u35<V>(input: V) -> V
where
    V: SimdFloat32,
    V::Engine: Simd<Vf32 = V>,
{
    portable::cos_u35(input)
}

#[inline(always)]
pub(crate) fn tan_u35<V>(input: V) -> V
where
    V: SimdFloat32,
    V::Engine: Simd<Vf32 = V>,
{
    portable::tan_u35(input)
}

#[inline(always)]
pub(crate) fn sinh_u35<V>(input: V) -> V
where
    V: SimdFloat32,
    V::Engine: Simd<Vf32 = V>,
{
    hyperbolic::sinh_u35(input)
}

#[inline(always)]
pub(crate) fn cosh_u35<V>(input: V) -> V
where
    V: SimdFloat32,
    V::Engine: Simd<Vf32 = V>,
{
    hyperbolic::cosh_u35(input)
}

#[inline(always)]
pub(crate) fn tanh_u35<V>(input: V) -> V
where
    V: SimdFloat32,
    V::Engine: Simd<Vf32 = V>,
{
    hyperbolic::tanh_u35(input)
}

#[inline(always)]
pub(crate) fn asinh_u35<V>(input: V) -> V
where
    V: SimdFloat32,
    V::Engine: Simd<Vf32 = V>,
{
    portable::asinh_u35(input)
}

#[inline(always)]
pub(crate) fn acosh_u35<V>(input: V) -> V
where
    V: SimdFloat32,
    V::Engine: Simd<Vf32 = V>,
{
    portable::acosh_u35(input)
}

#[inline(always)]
pub(crate) fn atanh_u35<V>(input: V) -> V
where
    V: SimdFloat32,
    V::Engine: Simd<Vf32 = V>,
{
    portable::atanh_u35(input)
}

#[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
#[inline(always)]
fn is_avx2_engine<E: Simd>() -> bool {
    core::any::TypeId::of::<E>() == core::any::TypeId::of::<crate::engines::avx2::Avx2>()
}

#[cfg(test)]
mod tests {
    #[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
    #[test]
    fn avx2_dispatch_gate_matches_only_avx2_engine() {
        use crate::engines::{avx2::Avx2, sse2::Sse2};

        assert!(super::is_avx2_engine::<Avx2>());
        assert!(!super::is_avx2_engine::<Sse2>());
    }
}