linear-srgb 0.6.12

Fast linear↔sRGB color space conversion with FMA acceleration and LUT support
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

//! sRGB transfer function — generic SIMD wrappers.
//!
//! Scalar sRGB delegates to `crate::rational_poly` (C0-continuous rational polynomial).
//! SIMD versions use the same polynomial coefficients inline.

use crate::rational_poly;

use magetypes::simd::backends::F32x4Convert;
use magetypes::simd::generic::f32x4;

#[allow(dead_code)]
#[inline(always)]
pub(crate) fn srgb_to_linear_x4<T: F32x4Convert>(t: T, v: f32x4<T>) -> f32x4<T> {
    let one = f32x4::splat(t, 1.0);
    let threshold = f32x4::splat(t, rational_poly::SRGB_THRESHOLD);
    let inv_12_92 = f32x4::splat(t, rational_poly::LINEAR_SCALE);

    let linear = v * inv_12_92;
    let poly =
        super::fast_math::eval_rational_poly_x4(t, v, rational_poly::S2L_P, rational_poly::S2L_Q)
            .min(one);

    let mask = v.simd_le(threshold);
    f32x4::blend(mask, linear, poly)
}

#[allow(dead_code)]
#[inline(always)]
pub(crate) fn linear_to_srgb_x4<T: F32x4Convert>(t: T, v: f32x4<T>) -> f32x4<T> {
    let one = f32x4::splat(t, 1.0);
    let threshold = f32x4::splat(t, rational_poly::LINEAR_THRESHOLD);
    let scale = f32x4::splat(t, rational_poly::TWELVE_92);

    let linear = v * scale;
    let s = v.sqrt();
    let poly =
        super::fast_math::eval_rational_poly_x4(t, s, rational_poly::L2S_P, rational_poly::L2S_Q)
            .min(one);

    let mask = v.simd_le(threshold);
    f32x4::blend(mask, linear, poly)
}

use magetypes::simd::backends::F32x8Convert;
use magetypes::simd::generic::f32x8;

#[inline(always)]
pub(crate) fn srgb_to_linear_x8<T: F32x8Convert>(t: T, v: f32x8<T>) -> f32x8<T> {
    let one = f32x8::splat(t, 1.0);
    let threshold = f32x8::splat(t, rational_poly::SRGB_THRESHOLD);
    let inv_12_92 = f32x8::splat(t, rational_poly::LINEAR_SCALE);

    let linear = v * inv_12_92;
    let poly =
        super::fast_math::eval_rational_poly_x8(t, v, rational_poly::S2L_P, rational_poly::S2L_Q)
            .min(one);

    let mask = v.simd_le(threshold);
    f32x8::blend(mask, linear, poly)
}

#[inline(always)]
pub(crate) fn linear_to_srgb_x8<T: F32x8Convert>(t: T, v: f32x8<T>) -> f32x8<T> {
    let one = f32x8::splat(t, 1.0);
    let threshold = f32x8::splat(t, rational_poly::LINEAR_THRESHOLD);
    let scale = f32x8::splat(t, rational_poly::TWELVE_92);

    let linear = v * scale;
    let s = v.sqrt();
    let poly =
        super::fast_math::eval_rational_poly_x8(t, s, rational_poly::L2S_P, rational_poly::L2S_Q)
            .min(one);

    let mask = v.simd_le(threshold);
    f32x8::blend(mask, linear, poly)
}

// =============================================================================
// Extended-range (sign-preserving, 6/6 rational polynomial) — x16
// =============================================================================

use magetypes::simd::backends::F32x16Backend;
use magetypes::simd::generic::f32x16;

/// Extended-range sRGB→linear, 16-wide. `sign(v) * eotf(|v|)` per CSS Color 4.
#[inline(always)]
pub(crate) fn srgb_to_linear_extended_x16<T: F32x16Backend>(t: T, v: f32x16<T>) -> f32x16<T> {
    use rational_poly::{EXT_S2L_P as P, EXT_S2L_Q as Q};
    let zero = f32x16::zero(t);
    let neg_mask = v.simd_lt(zero);
    let abs_v = v.abs();

    let linear_result = abs_v * f32x16::splat(t, rational_poly::LINEAR_SCALE);

    let x = abs_v;
    let yp = f32x16::splat(t, P[6]).mul_add(x, f32x16::splat(t, P[5]));
    let yp = yp.mul_add(x, f32x16::splat(t, P[4]));
    let yp = yp.mul_add(x, f32x16::splat(t, P[3]));
    let yp = yp.mul_add(x, f32x16::splat(t, P[2]));
    let yp = yp.mul_add(x, f32x16::splat(t, P[1]));
    let yp = yp.mul_add(x, f32x16::splat(t, P[0]));

    let yq = f32x16::splat(t, Q[6]).mul_add(x, f32x16::splat(t, Q[5]));
    let yq = yq.mul_add(x, f32x16::splat(t, Q[4]));
    let yq = yq.mul_add(x, f32x16::splat(t, Q[3]));
    let yq = yq.mul_add(x, f32x16::splat(t, Q[2]));
    let yq = yq.mul_add(x, f32x16::splat(t, Q[1]));
    let yq = yq.mul_add(x, f32x16::splat(t, Q[0]));

    let power_result = yp / yq;

    let thresh_mask = abs_v.simd_lt(f32x16::splat(t, rational_poly::SRGB_THRESHOLD));
    let result = f32x16::blend(thresh_mask, linear_result, power_result);
    f32x16::blend(neg_mask, -result, result)
}

/// Extended-range linear→sRGB, 16-wide. `sign(v) * oetf(|v|)` per CSS Color 4.
#[inline(always)]
pub(crate) fn linear_to_srgb_extended_x16<T: F32x16Backend>(t: T, v: f32x16<T>) -> f32x16<T> {
    use rational_poly::{EXT_L2S_P as P, EXT_L2S_Q as Q};
    let zero = f32x16::zero(t);
    let neg_mask = v.simd_lt(zero);
    let abs_v = v.abs();

    let linear_result = abs_v * f32x16::splat(t, rational_poly::TWELVE_92);

    let x = abs_v.sqrt();
    let yp = f32x16::splat(t, P[6]).mul_add(x, f32x16::splat(t, P[5]));
    let yp = yp.mul_add(x, f32x16::splat(t, P[4]));
    let yp = yp.mul_add(x, f32x16::splat(t, P[3]));
    let yp = yp.mul_add(x, f32x16::splat(t, P[2]));
    let yp = yp.mul_add(x, f32x16::splat(t, P[1]));
    let yp = yp.mul_add(x, f32x16::splat(t, P[0]));

    let yq = f32x16::splat(t, Q[6]).mul_add(x, f32x16::splat(t, Q[5]));
    let yq = yq.mul_add(x, f32x16::splat(t, Q[4]));
    let yq = yq.mul_add(x, f32x16::splat(t, Q[3]));
    let yq = yq.mul_add(x, f32x16::splat(t, Q[2]));
    let yq = yq.mul_add(x, f32x16::splat(t, Q[1]));
    let yq = yq.mul_add(x, f32x16::splat(t, Q[0]));

    let power_result = yp / yq;

    let thresh_mask = abs_v.simd_lt(f32x16::splat(t, rational_poly::LINEAR_THRESHOLD));
    let result = f32x16::blend(thresh_mask, linear_result, power_result);
    f32x16::blend(neg_mask, -result, result)
}