Unified GPU-accelerated FFT for Rust.

gpufft exposes a single trait surface that works the same across multiple GPU backends. Backends are selected at build time via Cargo features; at least one must be enabled.

Quick start

# #[cfg(feature = "vulkan")]
# fn demo() -> Result<(), Box<dyn std::error::Error>> {
use gpufft::{
    vulkan::VulkanBackend, BufferOps, C2cPlanOps, Device, Direction, PlanDesc, Shape,
};
use num_complex::Complex32;

let device = VulkanBackend::new_device(Default::default())?;
let mut buffer = device.alloc::<Complex32>(1024)?;
buffer.write(&vec![Complex32::default(); 1024])?;

let mut plan = device.plan_c2c::<Complex32>(&PlanDesc {
    shape: Shape::D1(1024),
    batch: 1,
    normalize: false,
})?;
plan.execute(&mut buffer, Direction::Forward)?;
# Ok(())
# }

Transform kinds

Each backend exposes three plan types, constructed through distinct device methods:

R2C output buffers and C2R input buffers are Hermitian-symmetric half-spectra (last dimension n / 2 + 1), matching VkFFT and cuFFT conventions. Use [Shape::complex_half_elements] to size them.

Buffers and plans are typed by their backend, so feeding a Vulkan buffer into a CUDA plan is a compile error.