yfft 0.1.0

Simple fast Fourier transform library optimized for small, power-of-two, single precision transforms and x86 systems 
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

//! Defines generic FFT kernels that don't make any assumptions about radix or instruction set.
//!
//! Performances
//! ------------
//!
//! According to a benchmark result, this kernel runs about 100x slower than a commercial-level FFT library on a Skylake
//! machine.

use super::{Kernel, KernelCreationParams, KernelParams, KernelType, SliceAccessor};

use num_complex::Complex;
use num_iter::range_step;
use num_traits::{One, Zero};

use super::super::{complex_from_slice, Num};

pub fn new_generic_kernel<T: 'static>(cparams: &KernelCreationParams) -> Box<Kernel<T>>
where
    T: Num,
{
    let full_circle = if cparams.inverse { 2 } else { -2 };
    let twiddle_delta = Complex::new(
        Zero::zero(),
        T::from(cparams.size / cparams.radix / cparams.unit).unwrap()
            * T::from(full_circle).unwrap()
            * T::PI()
            / T::from(cparams.size).unwrap(),
    )
    .exp();
    let coef_delta = Complex::new(
        Zero::zero(),
        T::from(full_circle).unwrap() * T::PI() / T::from(cparams.radix).unwrap(),
    )
    .exp();

    match cparams.kernel_type {
        KernelType::Dit => Box::new(GenericDitKernel {
            cparams: *cparams,
            twiddle_delta: twiddle_delta,
            coef_delta: coef_delta,
        }),
        KernelType::Dif => Box::new(GenericDifKernel {
            cparams: *cparams,
            twiddle_delta: twiddle_delta,
            coef_delta: coef_delta,
        }),
    }
}

#[derive(Debug)]
struct GenericDitKernel<T> {
    cparams: KernelCreationParams,
    twiddle_delta: Complex<T>,
    /// sub-FFT twiddle unit
    coef_delta: Complex<T>,
}

#[derive(Debug)]
struct GenericDifKernel<T> {
    cparams: KernelCreationParams,
    twiddle_delta: Complex<T>,
    /// sub-FFT twiddle unit
    coef_delta: Complex<T>,
}

impl<T> Kernel<T> for GenericDitKernel<T>
where
    T: Num,
{
    fn transform(&self, params: &mut KernelParams<T>) {
        let cparams = &self.cparams;
        let mut data = unsafe { SliceAccessor::new(&mut params.coefs[0..cparams.size * 2]) };
        let mut wa = unsafe { SliceAccessor::new(&mut params.work_area[0..cparams.radix * 2]) };

        let twiddle_delta = self.twiddle_delta;
        let coef_delta = self.coef_delta;

        for x in range_step(0, cparams.size, cparams.unit * cparams.radix) {
            let mut twiddle_1: Complex<T> = Complex::one();
            for y in 0..cparams.unit {
                for z in 0..cparams.radix {
                    wa[z * 2] = data[(x + y + z * cparams.unit) * 2];
                    wa[z * 2 + 1] = data[(x + y + z * cparams.unit) * 2 + 1];
                }
                let mut coef_1: Complex<T> = Complex::one();
                for z in 0..cparams.radix {
                    let mut c: Complex<T> = Complex::zero();
                    let mut coef_2: Complex<T> = Complex::one();
                    let coef_1_tw: Complex<T> = coef_1 * twiddle_1;
                    for w in 0..cparams.radix {
                        c = c + coef_2 * complex_from_slice(&wa[w * 2..]);
                        coef_2 = coef_2 * coef_1_tw;
                    }
                    data[(x + y + z * cparams.unit) * 2] = c.re;
                    data[(x + y + z * cparams.unit) * 2 + 1] = c.im;
                    coef_1 = coef_1 * coef_delta;
                }
                twiddle_1 = twiddle_1 * twiddle_delta;
            }
        }
    }

    fn required_work_area_size(&self) -> usize {
        self.cparams.radix * 2
    }
}

impl<T> Kernel<T> for GenericDifKernel<T>
where
    T: Num,
{
    fn transform(&self, params: &mut KernelParams<T>) {
        let cparams = &self.cparams;
        let mut data = unsafe { SliceAccessor::new(&mut params.coefs[0..cparams.size * 2]) };
        let mut wa = unsafe { SliceAccessor::new(&mut params.work_area[0..cparams.radix * 2]) };

        let twiddle_delta = self.twiddle_delta;
        let coef_delta = self.coef_delta;

        for x in range_step(0, cparams.size, cparams.unit * cparams.radix) {
            let mut twiddle_1: Complex<T> = Complex::one();
            for y in 0..cparams.unit {
                for z in 0..cparams.radix {
                    wa[z * 2] = data[(x + y + z * cparams.unit) * 2];
                    wa[z * 2 + 1] = data[(x + y + z * cparams.unit) * 2 + 1];
                }
                let mut twiddle_2: Complex<T> = Complex::one();
                let mut coef_1: Complex<T> = Complex::one();
                for z in 0..cparams.radix {
                    let mut c: Complex<T> = Complex::zero();
                    let mut coef_2 = twiddle_2;
                    for w in 0..cparams.radix {
                        c = c + coef_2 * complex_from_slice(&wa[w * 2..]);
                        coef_2 = coef_2 * coef_1;
                    }
                    data[(x + y + z * cparams.unit) * 2] = c.re;
                    data[(x + y + z * cparams.unit) * 2 + 1] = c.im;
                    twiddle_2 = twiddle_2 * twiddle_1;
                    coef_1 = coef_1 * coef_delta;
                }
                twiddle_1 = twiddle_1 * twiddle_delta;
            }
        }
    }

    fn required_work_area_size(&self) -> usize {
        self.cparams.radix * 2
    }
}