gpu-fft 1.2.0

A Rust library for performing Fast Fourier Transform (FFT) and Inverse FFT using GPU acceleration.
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

use std::f32::consts::PI;
use std::time::Duration;

use criterion::{BatchSize, BenchmarkId, Criterion, Throughput, criterion_group, criterion_main};

type WgpuRuntime = cubecl::wgpu::WgpuRuntime;

const SIZES: &[usize] = &[256, 1_024, 4_096, 16_384, 65_536];

fn sine_wave(n: usize) -> Vec<f32> {
    (0..n)
        .map(|i| (2.0 * PI * i as f32 / n as f32).sin())
        .collect()
}

// ── Compare forward FFT ─────────────────────────────────────────────────────

fn bench_compare_fft(c: &mut Criterion) {
    let device = cubecl::wgpu::WgpuDevice::default();
    let mut group = c.benchmark_group("compare_fft");
    group.warm_up_time(Duration::from_secs(2));
    group.measurement_time(Duration::from_secs(5));

    for &n in SIZES {
        let input = sine_wave(n);
        group.throughput(Throughput::Elements(n as u64));

        group.bench_with_input(BenchmarkId::new("wgpu", n), &input, |b, input| {
            b.iter_batched(
                || input.clone(),
                |inp| gpu_fft::fft::fft::<WgpuRuntime>(&device, &inp),
                BatchSize::SmallInput,
            );
        });

        group.bench_with_input(BenchmarkId::new("mlx", n), &input, |b, input| {
            b.iter_batched(
                || input.clone(),
                |inp| gpu_fft::mlx::fft::fft(&inp),
                BatchSize::SmallInput,
            );
        });
    }

    group.finish();
}

// ── Compare inverse FFT ─────────────────────────────────────────────────────

fn bench_compare_ifft(c: &mut Criterion) {
    let device = cubecl::wgpu::WgpuDevice::default();
    let mut group = c.benchmark_group("compare_ifft");
    group.warm_up_time(Duration::from_secs(2));
    group.measurement_time(Duration::from_secs(5));

    for &n in SIZES {
        let (real, imag) = gpu_fft::fft::fft::<WgpuRuntime>(&device, &sine_wave(n));
        group.throughput(Throughput::Elements(n as u64));

        group.bench_with_input(
            BenchmarkId::new("wgpu", n),
            &(real.clone(), imag.clone()),
            |b, (real, imag)| {
                b.iter_batched(
                    || (real.clone(), imag.clone()),
                    |(re, im)| gpu_fft::ifft::ifft::<WgpuRuntime>(&device, &re, &im),
                    BatchSize::SmallInput,
                );
            },
        );

        group.bench_with_input(
            BenchmarkId::new("mlx", n),
            &(real.clone(), imag.clone()),
            |b, (real, imag)| {
                b.iter_batched(
                    || (real.clone(), imag.clone()),
                    |(re, im)| gpu_fft::mlx::fft::ifft(&re, &im),
                    BatchSize::SmallInput,
                );
            },
        );
    }

    group.finish();
}

// ── Compare round-trip (FFT → IFFT) ─────────────────────────────────────────

fn bench_compare_roundtrip(c: &mut Criterion) {
    let device = cubecl::wgpu::WgpuDevice::default();
    let mut group = c.benchmark_group("compare_roundtrip");
    group.warm_up_time(Duration::from_secs(2));
    group.measurement_time(Duration::from_secs(5));

    for &n in SIZES {
        let input = sine_wave(n);
        group.throughput(Throughput::Elements(n as u64));

        group.bench_with_input(BenchmarkId::new("wgpu", n), &input, |b, input| {
            b.iter_batched(
                || input.clone(),
                |inp| {
                    let (re, im) = gpu_fft::fft::fft::<WgpuRuntime>(&device, &inp);
                    gpu_fft::ifft::ifft::<WgpuRuntime>(&device, &re, &im)
                },
                BatchSize::SmallInput,
            );
        });

        group.bench_with_input(BenchmarkId::new("mlx", n), &input, |b, input| {
            b.iter_batched(
                || input.clone(),
                |inp| {
                    let (re, im) = gpu_fft::mlx::fft::fft(&inp);
                    gpu_fft::mlx::fft::ifft(&re, &im)
                },
                BatchSize::SmallInput,
            );
        });
    }

    group.finish();
}

criterion_group!(
    benches,
    bench_compare_fft,
    bench_compare_ifft,
    bench_compare_roundtrip,
);
criterion_main!(benches);