yum/vfft

git clone https://git.yummers.dev/yum/vfft

yumVersion bumpabf4a25

master
branchesmasterall refs →
1.2 KiB49 linesraw

vfft

vfft is a very fast fourier transform implementation. It currently optimizes for small, power-of-two inputs sizes.

vfft is under heavy development. If you are in need of a high-quality CPU-based FFT, use rustfft. If you need a GPU-based one, use VkFFT.

The goal of this project is to define simple, state-of-the-art fft implementations for small input sizes (under ~8k). My current north star is to meet or exceed VkFft's peformance for workloads where the FFT input and output remains resident on the GPU. This is useful for realtime audio-reactive graphics and ocean wave simulations, and probably more. Eventually, I would also like to implement a world-class CPU implementation, but this is a secondary concern.

Portability

MacOS and Linux are confirmed to work. Windows not tested.

API

use num::Complex;
use vfft::cpu::{ExecutionPlan, fft};
...
let mut data: Vec<Complex<f32>> = ...;
let mut plan = ExecutionPlan::new(data)?;
fft(&mut result_fft, &mut plan);

Testing

# Accuracy
$ cargo test -- --no-capture
# Performance
$ cargo bench

AI Disclosure

This source code was written by a human and checked for errors with an LLM.

License.

MIT License.

1# vfft
2
3vfft is a very fast fourier transform implementation. It currently optimizes
4for small, power-of-two inputs sizes.
5
6**vfft is under heavy development. If you are in need of a high-quality CPU-based
7FFT, use rustfft. If you need a GPU-based one, use VkFFT.**
8
9The goal of this project is to define simple, state-of-the-art fft
10implementations for small input sizes (under ~8k). My current north star is to
11meet or exceed VkFft's peformance for workloads where the FFT input and output
12remains resident on the GPU. This is useful for realtime audio-reactive
13graphics and ocean wave simulations, and probably more. Eventually, I would
14also like to implement a world-class CPU implementation, but this is a
15secondary concern.
16
17## Portability
18
19MacOS and Linux are confirmed to work. Windows not tested.
20
21## API
22
23```rust
24use num::Complex;
25use vfft::cpu::{ExecutionPlan, fft};
26...
27let mut data: Vec<Complex<f32>> = ...;
28let mut plan = ExecutionPlan::new(data)?;
29fft(&mut result_fft, &mut plan);
30```
31
32## Testing
33
34```bash
35# Accuracy
36$ cargo test -- --no-capture
37# Performance
38$ cargo bench
39```
40
41## AI Disclosure
42
43This source code was written by a human and checked for errors with an
44LLM.
45
46## License.
47
48MIT License.
49