dtfft-cuda12x-openmpi 3.2.0

Python bindings for the dtFFT library — distributed FFT via MPI

dtfft-cuda12x — CPU + GPU build for CUDA 12.x [OpenMPI]

Python bindings for dtFFT — a high-performance library for parallel data transpositions and Fast Fourier Transforms via MPI.

This is the dtfft-cuda12x-openmpi variant. GPU-accelerated transforms via cuFFT on NVIDIA GPUs with CUDA 12.x.

Included backends

Backend	Available
Transpose (MPI)	✓
FFTW3	—
cuFFT	✓

Installation

pip install dtfft-cuda12x-openmpi

Note: cuFFT and a CUDA 12.x toolkit must be present on the system. CuPy is installed automatically as a dependency.

Requirements

• MPI runtime (OpenMPI or MPICH)
• Python ≥ 3.9
• CUDA 12.x toolkit
• NVIDIA GPU (Volta or newer)

Quick start

import numpy as np
from mpi4py import MPI
import dtfft

# Create a 3-D complex-to-complex plan
plan = dtfft.PlanC2C([256, 256, 256], comm=MPI.COMM_WORLD)

# Allocate MPI-decomposed buffers
x = plan.get_ndarray(plan.alloc_size)
y = plan.get_ndarray(plan.alloc_size)

x[...] = np.random.random(x.shape) + 1j * np.random.random(x.shape)

# Forward transform  (pencil decomposition applied automatically)
plan.execute(x, y, dtfft.Execute.FORWARD)

# Backward transform
plan.execute(y, x, dtfft.Execute.BACKWARD)

Package variants

Multiple PyPI distributions are provided so you can declare an exact, conflict-free dependency for your environment:

PyPI package	FFT backend	MPI	Platform	Extra deps
dtfft-openmpi	none (transpose-only)	OpenMPI	CPU	—
dtfft-mpich	none (transpose-only)	MPICH	CPU	—
dtfft-fftw-openmpi	FFTW3	OpenMPI	CPU	system libfftw3
dtfft-fftw-mpich	FFTW3	MPICH	CPU	system libfftw3
dtfft-cuda12x-openmpi	cuFFT	OpenMPI	CPU + NVIDIA GPU (CUDA 12)	cupy-cuda12x
dtfft-cuda12x-mpich	cuFFT	MPICH	CPU + NVIDIA GPU (CUDA 12)	cupy-cuda12x

All packages share the same importable namespace: import dtfft.

Documentation

• API reference
• Building from source
• Examples

License

GPL v3 — see LICENSE.