Crate wgsl_fft 

GPU-accelerated FFT using wgpu compute shaders.

This crate provides fast Fourier transform (FFT) computation on GPU hardware using WebGPU compute shaders via the wgpu library.

Features

• GPU-accelerated: Uses wgpu compute shaders for high-performance FFT computation.
• Stockham Radix-4/2: Fast algorithm for power-of-two FFT sizes.
• Arbitrary sizes: Bluestein’s algorithm for non-power-of-two sizes (GPU-accelerated via power-of-2 FFTs).
• Batch processing: Efficiently process multiple FFTs in a single call.
• Fallback support: Automatically falls back to CPU software rendering if no GPU is available.

Quick Start

use wgsl_fft::GpuFft;
use num_complex::Complex;

let fft = GpuFft::new().expect("GPU or CPU fallback required");
let input = vec![vec![Complex::new(1.0, 0.0); 1024]];
let spectrum = fft.fft(&input).expect("FFT failed");

Module Structure

• [fft] - Main FFT implementation with GpuFft and FftExecutor trait
• [pipelines] - Pre-compiled FFT pipelines for embedding in larger GPU pipelines
• shaders - WGSL compute shader source code
• benchmark - Benchmarking utilities
• [error] - Error types for FFT operations

Modules

benchmark

rivals

The FFT Rivalry Registry

shaders

Structs

FftPipelines

Pre-compiled Cooley-Tukey Radix-2 FFT pipelines for embedding in a larger GPU pipeline.

FftUniforms

Uniforms passed to the compute shader (16-byte aligned).

GpuFft

GPU-accelerated FFT engine backed by wgpu compute shaders.

SizeCache

Pre-allocated GPU resources for a specific FFT size.

Enums

FftDirection

Direction for FFT operations

FftError

Main error type for FFT operations

Traits

FftExecutor

Trait for FFT implementations that can be benchmarked.

GpuFftTrait

Trait for GPU FFT implementations that support GPU-only benchmarking.

Type Aliases

Result

Result type alias for FFT operations