oxifft-codegen-impl 0.3.1

Internal codelet generation logic for OxiFFT (not a proc-macro; used by oxifft-codegen and benchmarks)
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

//! Mirror of `oxifft::dft::codelets::winograd_constants` for use by codegen-impl.
//!
//! Values must stay in sync — run the cross-validation test to verify.
//!
//! All constants are f64 precision, computed from exact expressions
//! cos(2πk/N) and sin(2πk/N). The forward DFT convention used throughout
//! is `W_N` = e^{-2πi/N}, so:
//!   - real parts: cos(2πk/N)
//!   - imaginary parts (for the negative sign): -sin(2πk/N)

// ─── DFT-3 ───────────────────────────────────────────────────────────────────
// cos(2π/3) = -1/2,  sin(2π/3) = √3/2

/// cos(2π/3) = −1/2
pub const C3_1: f64 = -0.5_f64;
/// sin(2π/3) = √3/2
pub const C3_2: f64 = 0.866_025_403_784_438_7_f64;

// ─── DFT-5 ───────────────────────────────────────────────────────────────────
// cos(2πk/5) and sin(2πk/5) for k = 1, 2

/// cos(2π/5)
pub const C5_COS1: f64 = 0.309_016_994_374_947_45_f64;
/// cos(4π/5)
pub const C5_COS2: f64 = -0.809_016_994_374_947_3_f64;
/// sin(2π/5)
pub const C5_SIN1: f64 = 0.951_056_516_295_153_5_f64;
/// sin(4π/5)
pub const C5_SIN2: f64 = 0.587_785_252_292_473_2_f64;

// ─── DFT-7 ───────────────────────────────────────────────────────────────────
// cos(2πk/7) and sin(2πk/7) for k = 1, 2, 3

/// cos(2π/7)
pub const C7_COS1: f64 = 0.623_489_801_858_733_6_f64;
/// cos(4π/7)
pub const C7_COS2: f64 = -0.222_520_933_956_314_34_f64;
/// cos(6π/7)
pub const C7_COS3: f64 = -0.900_968_867_902_419_f64;
/// sin(2π/7)
pub const C7_SIN1: f64 = 0.781_831_482_468_029_8_f64;
/// sin(4π/7)
pub const C7_SIN2: f64 = 0.974_927_912_181_823_6_f64;
/// sin(6π/7)
pub const C7_SIN3: f64 = 0.433_883_739_117_558_23_f64;

// ─── DFT-9 ───────────────────────────────────────────────────────────────────
// cos(2πk/9) and sin(2πk/9) for k = 1, 2, 3, 4

/// cos(2π/9)
pub const C9_COS1: f64 = 0.766_044_443_118_978_f64;
/// cos(4π/9)
pub const C9_COS2: f64 = 0.173_648_177_666_930_41_f64;
/// cos(6π/9) = cos(2π/3) = -1/2
pub const C9_COS3: f64 = -0.5_f64;
/// cos(8π/9)
pub const C9_COS4: f64 = -0.939_692_620_785_908_3_f64;
/// sin(2π/9)
pub const C9_SIN1: f64 = 0.642_787_609_686_539_3_f64;
/// sin(4π/9)
pub const C9_SIN2: f64 = 0.984_807_753_012_208_f64;
/// sin(6π/9) = sin(2π/3) = √3/2
pub const C9_SIN3: f64 = 0.866_025_403_784_438_7_f64;
/// sin(8π/9)
pub const C9_SIN4: f64 = 0.342_020_143_325_668_9_f64;

// ─── DFT-11 ──────────────────────────────────────────────────────────────────
// cos(2πk/11) and sin(2πk/11) for k = 1..5

/// cos(2π/11)
pub const C11_COS1: f64 = 0.841_253_532_831_181_2_f64;
/// cos(4π/11)
pub const C11_COS2: f64 = 0.415_415_013_001_886_44_f64;
/// cos(6π/11)
pub const C11_COS3: f64 = -0.142_314_838_273_285_f64;
/// cos(8π/11)
pub const C11_COS4: f64 = -0.654_860_733_945_285_1_f64;
/// cos(10π/11)
pub const C11_COS5: f64 = -0.959_492_973_614_497_4_f64;
/// sin(2π/11)
pub const C11_SIN1: f64 = 0.540_640_817_455_597_6_f64;
/// sin(4π/11)
pub const C11_SIN2: f64 = 0.909_631_995_354_518_3_f64;
/// sin(6π/11)
pub const C11_SIN3: f64 = 0.989_821_441_880_932_8_f64;
/// sin(8π/11)
pub const C11_SIN4: f64 = 0.755_749_574_354_258_3_f64;
/// sin(10π/11)
pub const C11_SIN5: f64 = 0.281_732_556_841_429_67_f64;

// ─── DFT-13 ──────────────────────────────────────────────────────────────────
// cos(2πk/13) and sin(2πk/13) for k = 1..6

/// cos(2π/13)
pub const C13_COS1: f64 = 0.885_456_025_653_209_9_f64;
/// cos(4π/13)
pub const C13_COS2: f64 = 0.568_064_746_731_155_8_f64;
/// cos(6π/13)
pub const C13_COS3: f64 = 0.120_536_680_255_323_f64;
/// cos(8π/13)
pub const C13_COS4: f64 = -0.354_604_887_042_535_45_f64;
/// cos(10π/13)
pub const C13_COS5: f64 = -0.748_510_748_171_101_2_f64;
/// cos(12π/13)
pub const C13_COS6: f64 = -0.970_941_817_426_052_f64;
/// sin(2π/13)
pub const C13_SIN1: f64 = 0.464_723_172_043_768_5_f64;
/// sin(4π/13)
pub const C13_SIN2: f64 = 0.822_983_865_893_656_4_f64;
/// sin(6π/13)
pub const C13_SIN3: f64 = 0.992_708_874_098_054_f64;
/// sin(8π/13)
pub const C13_SIN4: f64 = 0.935_016_242_685_414_8_f64;
/// sin(10π/13)
pub const C13_SIN5: f64 = 0.663_122_658_240_795_2_f64;
/// sin(12π/13)
pub const C13_SIN6: f64 = 0.239_315_664_287_557_68_f64;

// ─── Cross-validation test ────────────────────────────────────────────────────

#[cfg(test)]
pub(crate) fn verify_constants_match_runtime() {
    // Verify that this mirror matches the runtime winograd_constants exactly.
    // These are the same values — just check they agree within f64 precision.
    let tol = 1e-13;
    let two_pi = 2.0 * std::f64::consts::PI;

    // DFT-3
    assert!((C3_1 - f64::cos(two_pi / 3.0)).abs() < tol, "C3_1");
    assert!((C3_2 - f64::sin(two_pi / 3.0)).abs() < tol, "C3_2");

    // DFT-5
    assert!((C5_COS1 - f64::cos(two_pi / 5.0)).abs() < tol, "C5_COS1");
    assert!(
        (C5_COS2 - f64::cos(2.0 * two_pi / 5.0)).abs() < tol,
        "C5_COS2"
    );
    assert!((C5_SIN1 - f64::sin(two_pi / 5.0)).abs() < tol, "C5_SIN1");
    assert!(
        (C5_SIN2 - f64::sin(2.0 * two_pi / 5.0)).abs() < tol,
        "C5_SIN2"
    );

    // DFT-7
    assert!((C7_COS1 - f64::cos(two_pi / 7.0)).abs() < tol, "C7_COS1");
    assert!(
        (C7_COS2 - f64::cos(2.0 * two_pi / 7.0)).abs() < tol,
        "C7_COS2"
    );
    assert!(
        (C7_COS3 - f64::cos(3.0 * two_pi / 7.0)).abs() < tol,
        "C7_COS3"
    );
    assert!((C7_SIN1 - f64::sin(two_pi / 7.0)).abs() < tol, "C7_SIN1");
    assert!(
        (C7_SIN2 - f64::sin(2.0 * two_pi / 7.0)).abs() < tol,
        "C7_SIN2"
    );
    assert!(
        (C7_SIN3 - f64::sin(3.0 * two_pi / 7.0)).abs() < tol,
        "C7_SIN3"
    );
}