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
148
149
150
151
152
153
154
155

//! BT.709 transfer function (scalar + generic SIMD).

use super::fast_math;

const BT709_ALPHA: f32 = 0.09929682680944;
const BT709_BETA: f32 = 0.018053968510807;

/// BT.709 EOTF: encoded → linear. Uses fast_powf, max error ~3e-6.
#[inline(always)]
pub fn bt709_to_linear(v: f32) -> f32 {
    if v < 4.5 * BT709_BETA {
        v / 4.5
    } else {
        fast_math::fast_powf((v + BT709_ALPHA) / (1.0 + BT709_ALPHA), 1.0 / 0.45)
    }
}

/// BT.709 inverse EOTF: linear → encoded. Uses fast_powf, max error ~3e-5.
#[inline(always)]
pub fn linear_to_bt709(v: f32) -> f32 {
    if v < BT709_BETA {
        4.5 * v
    } else {
        (1.0 + BT709_ALPHA) * fast_math::fast_powf(v, 0.45) - BT709_ALPHA
    }
}

// =============================================================================
// Generic SIMD — x4
// =============================================================================

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

#[allow(dead_code)]
#[inline(always)]
pub(crate) fn bt709_to_linear_x4<T: F32x4Convert>(t: T, v: f32x4<T>) -> f32x4<T> {
    let threshold = f32x4::splat(t, 4.5 * BT709_BETA);
    let inv_4_5 = f32x4::splat(t, 1.0 / 4.5);
    let alpha = f32x4::splat(t, BT709_ALPHA);
    let one_plus_alpha = f32x4::splat(t, 1.0 + BT709_ALPHA);

    let linear = v * inv_4_5;

    // No MIN_POSITIVE clamp on `normalized`: `power` is always blended away
    // when v < threshold, and fast_powf never faults on negative input
    // (all bit-manipulation, finite output), so garbage from out-of-domain
    // lanes never escapes.
    let normalized = (v + alpha) / one_plus_alpha;
    let power = fast_math::fast_powf_x4(t, normalized, 1.0 / 0.45);

    let mask = v.simd_lt(threshold);
    f32x4::blend(mask, linear, power)
}

#[allow(dead_code)]
#[inline(always)]
pub(crate) fn linear_to_bt709_x4<T: F32x4Convert>(t: T, v: f32x4<T>) -> f32x4<T> {
    let threshold = f32x4::splat(t, BT709_BETA);
    let scale_4_5 = f32x4::splat(t, 4.5);
    let one_plus_alpha = f32x4::splat(t, 1.0 + BT709_ALPHA);
    let alpha = f32x4::splat(t, BT709_ALPHA);

    let linear = v * scale_4_5;

    // See bt709_to_linear_x4: no MIN_POSITIVE clamp needed; `power` is
    // blended away for v < threshold and fast_powf is branch-free.
    let power = fast_math::fast_powf_x4(t, v, 0.45);
    let power = one_plus_alpha.mul_add(power, -alpha);

    let mask = v.simd_lt(threshold);
    f32x4::blend(mask, linear, power)
}

// =============================================================================
// Generic SIMD — x8
// =============================================================================

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

#[inline(always)]
pub(crate) fn bt709_to_linear_x8<T: F32x8Convert>(t: T, v: f32x8<T>) -> f32x8<T> {
    let threshold = f32x8::splat(t, 4.5 * BT709_BETA);
    let inv_4_5 = f32x8::splat(t, 1.0 / 4.5);
    let alpha = f32x8::splat(t, BT709_ALPHA);
    let one_plus_alpha = f32x8::splat(t, 1.0 + BT709_ALPHA);

    let linear = v * inv_4_5;

    // See bt709_to_linear_x4: no MIN_POSITIVE clamp needed.
    let normalized = (v + alpha) / one_plus_alpha;
    let power = fast_math::fast_powf_x8(t, normalized, 1.0 / 0.45);

    let mask = v.simd_lt(threshold);
    f32x8::blend(mask, linear, power)
}

#[inline(always)]
pub(crate) fn linear_to_bt709_x8<T: F32x8Convert>(t: T, v: f32x8<T>) -> f32x8<T> {
    let threshold = f32x8::splat(t, BT709_BETA);
    let scale_4_5 = f32x8::splat(t, 4.5);
    let one_plus_alpha = f32x8::splat(t, 1.0 + BT709_ALPHA);
    let alpha = f32x8::splat(t, BT709_ALPHA);

    let linear = v * scale_4_5;

    // See linear_to_bt709_x4: no MIN_POSITIVE clamp needed.
    let power = fast_math::fast_powf_x8(t, v, 0.45);
    let power = one_plus_alpha.mul_add(power, -alpha);

    let mask = v.simd_lt(threshold);
    f32x8::blend(mask, linear, power)
}

// =============================================================================
// Generic SIMD — x16
// =============================================================================

use magetypes::simd::backends::F32x16Convert;
use magetypes::simd::generic::f32x16;

#[inline(always)]
pub(crate) fn bt709_to_linear_x16<T: F32x16Convert>(t: T, v: f32x16<T>) -> f32x16<T> {
    let threshold = f32x16::splat(t, 4.5 * BT709_BETA);
    let inv_4_5 = f32x16::splat(t, 1.0 / 4.5);
    let alpha = f32x16::splat(t, BT709_ALPHA);
    let one_plus_alpha = f32x16::splat(t, 1.0 + BT709_ALPHA);

    let linear = v * inv_4_5;

    // See bt709_to_linear_x4: no MIN_POSITIVE clamp needed.
    let normalized = (v + alpha) / one_plus_alpha;
    let power = fast_math::fast_powf_x16(t, normalized, 1.0 / 0.45);

    let mask = v.simd_lt(threshold);
    f32x16::blend(mask, linear, power)
}

#[inline(always)]
pub(crate) fn linear_to_bt709_x16<T: F32x16Convert>(t: T, v: f32x16<T>) -> f32x16<T> {
    let threshold = f32x16::splat(t, BT709_BETA);
    let scale_4_5 = f32x16::splat(t, 4.5);
    let one_plus_alpha = f32x16::splat(t, 1.0 + BT709_ALPHA);
    let alpha = f32x16::splat(t, BT709_ALPHA);

    let linear = v * scale_4_5;

    // See linear_to_bt709_x4: no MIN_POSITIVE clamp needed.
    let power = fast_math::fast_powf_x16(t, v, 0.45);
    let power = one_plus_alpha.mul_add(power, -alpha);

    let mask = v.simd_lt(threshold);
    f32x16::blend(mask, linear, power)
}