linear-srgb 0.7.0

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
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213

//! PQ (SMPTE ST 2084) transfer function (scalar + generic SIMD).

use super::fast_math;
#[allow(unused_imports)]
use num_traits::Float; // provides powf/sqrt via libm in no_std

// =============================================================================
// Scalar
// =============================================================================

/// PQ EOTF: signal → linear. Two-range rational polynomial, sub-U16 roundtrip.
///
/// Zero `powf()` calls for signal > 0.02 — uses `x + x*x` input transformation
/// with separate polynomial fits for low and high signal ranges.
/// Very small signal values use exact formula.
#[inline(always)]
pub fn pq_to_linear(v: f32) -> f32 {
    if v <= 0.0 {
        return 0.0;
    }
    if v < 0.02 {
        return pq_to_linear_exact(v);
    }
    let x = v + v * v;
    if v < 0.25 {
        fast_math::eval_rational_poly(x, PQ_EOTF_P_SMALL, PQ_EOTF_Q_SMALL)
    } else {
        fast_math::eval_rational_poly(x, PQ_EOTF_P_LARGE, PQ_EOTF_Q_LARGE)
    }
}

/// PQ inverse EOTF: linear → signal. Rational polynomial on x^(1/4), max error ~3e-6.
///
/// Two-range approximation. Very small values (v < 1e-4) use exact powf.
/// Coefficients from libjxl.
#[inline(always)]
pub fn linear_to_pq(v: f32) -> f32 {
    if v <= 0.0 {
        return 0.0;
    }
    if v < 1e-4 {
        return pq_from_linear_exact(v);
    }
    let a = v.sqrt().sqrt();
    if a < 0.1 {
        fast_math::eval_rational_poly(a, PQ_INV_P_SMALL, PQ_INV_Q_SMALL)
    } else {
        fast_math::eval_rational_poly(a, PQ_INV_P_LARGE, PQ_INV_Q_LARGE)
    }
}

#[inline(always)]
fn pq_to_linear_exact(v: f32) -> f32 {
    const M1: f32 = 0.1593017578125;
    const M2: f32 = 78.84375;
    const C1: f32 = 0.8359375;
    const C2: f32 = 18.8515625;
    const C3: f32 = 18.6875;
    let vp = v.powf(1.0 / M2);
    let num = (vp - C1).max(0.0);
    let den = C2 - C3 * vp;
    if den <= 0.0 {
        return 1.0;
    }
    (num / den).powf(1.0 / M1)
}

#[inline(always)]
fn pq_from_linear_exact(v: f32) -> f32 {
    const M1: f32 = 0.1593017578125;
    const M2: f32 = 78.84375;
    const C1: f32 = 0.8359375;
    const C2: f32 = 18.8515625;
    const C3: f32 = 18.6875;
    let vp = v.powf(M1);
    let num = C1 + C2 * vp;
    let den = 1.0 + C3 * vp;
    (num / den).powf(M2)
}

// =============================================================================
// Coefficients
// =============================================================================

// Two-range EOTF: split at signal = 0.25 for sub-U16-step roundtrip accuracy.
// Fitted with relative-error weighting (minimax in signal-space roundtrip).
pub(crate) const PQ_EOTF_P_SMALL: [f32; 5] = [
    4.36033203e-10,
    -1.53651077e-07,
    2.06864196e-06,
    2.80304711e-03,
    1.47809948e-02,
];
pub(crate) const PQ_EOTF_Q_SMALL: [f32; 5] = [
    -9.89955022e-03,
    1.89911767e+00,
    -1.34828020e+00,
    -7.10547502e-01,
    1.00000000e+00,
];
pub(crate) const PQ_EOTF_P_LARGE: [f32; 5] = [
    5.43046166e-04,
    -6.31017431e-03,
    2.90480734e-01,
    9.80363032e-01,
    2.22819483e-01,
];
pub(crate) const PQ_EOTF_Q_LARGE: [f32; 5] = [
    1.58688561e+02,
    -1.60764060e+02,
    6.51370372e+01,
    -1.26438391e+01,
    1.00000000e+00,
];

pub(crate) const PQ_INV_P_LARGE: [f32; 5] =
    [1.351392e-2, -1.095778, 5.522776e1, 1.492516e2, 4.838434e1];
pub(crate) const PQ_INV_Q_LARGE: [f32; 5] =
    [1.012416, 2.016708e1, 9.26371e1, 1.120607e2, 2.590418e1];

pub(crate) const PQ_INV_P_SMALL: [f32; 5] = [
    9.863406e-6,
    3.881234e-1,
    1.352821e2,
    6.889862e4,
    -2.864824e5,
];
pub(crate) const PQ_INV_Q_SMALL: [f32; 5] =
    [3.371868e1, 1.477719e3, 1.608477e4, -4.389884e4, -2.072546e5];

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

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

#[allow(dead_code)]
#[inline(always)]
pub(crate) fn pq_to_linear_x4<T: F32x4Convert>(t: T, v: f32x4<T>) -> f32x4<T> {
    let zero = f32x4::zero(t);
    let a = v.max(zero);
    let x = a.mul_add(a, a);

    let threshold = f32x4::splat(t, 0.25);
    let large = fast_math::eval_rational_poly_x4(t, x, PQ_EOTF_P_LARGE, PQ_EOTF_Q_LARGE);
    let small = fast_math::eval_rational_poly_x4(t, x, PQ_EOTF_P_SMALL, PQ_EOTF_Q_SMALL);

    let mask = a.simd_lt(threshold);
    let result = f32x4::blend(mask, small, large);

    let pos_mask = v.simd_gt(zero);
    result & pos_mask
}

#[allow(dead_code)]
#[inline(always)]
pub(crate) fn linear_to_pq_x4<T: F32x4Convert>(t: T, v: f32x4<T>) -> f32x4<T> {
    let zero = f32x4::zero(t);
    let a = v.max(zero);
    let a4 = a.sqrt().sqrt();

    let threshold = f32x4::splat(t, 0.1);
    let large = fast_math::eval_rational_poly_x4(t, a4, PQ_INV_P_LARGE, PQ_INV_Q_LARGE);
    let small = fast_math::eval_rational_poly_x4(t, a4, PQ_INV_P_SMALL, PQ_INV_Q_SMALL);

    let mask = a4.simd_lt(threshold);
    let result = f32x4::blend(mask, small, large);

    let pos_mask = v.simd_gt(zero);
    result & pos_mask
}

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

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

#[inline(always)]
pub(crate) fn pq_to_linear_x8<T: F32x8Convert>(t: T, v: f32x8<T>) -> f32x8<T> {
    let zero = f32x8::zero(t);
    let a = v.max(zero);
    let x = a.mul_add(a, a);

    let threshold = f32x8::splat(t, 0.25);
    let large = fast_math::eval_rational_poly_x8(t, x, PQ_EOTF_P_LARGE, PQ_EOTF_Q_LARGE);
    let small = fast_math::eval_rational_poly_x8(t, x, PQ_EOTF_P_SMALL, PQ_EOTF_Q_SMALL);

    let mask = a.simd_lt(threshold);
    let result = f32x8::blend(mask, small, large);

    let pos_mask = v.simd_gt(zero);
    result & pos_mask
}

#[inline(always)]
pub(crate) fn linear_to_pq_x8<T: F32x8Convert>(t: T, v: f32x8<T>) -> f32x8<T> {
    let zero = f32x8::zero(t);
    let a = v.max(zero);
    let a4 = a.sqrt().sqrt();

    let threshold = f32x8::splat(t, 0.1);
    let large = fast_math::eval_rational_poly_x8(t, a4, PQ_INV_P_LARGE, PQ_INV_Q_LARGE);
    let small = fast_math::eval_rational_poly_x8(t, a4, PQ_INV_P_SMALL, PQ_INV_Q_SMALL);

    let mask = a4.simd_lt(threshold);
    let result = f32x8::blend(mask, small, large);

    let pos_mask = v.simd_gt(zero);
    result & pos_mask
}