How to perform a 2D FFT on an image using PocketFFT?

Prepare your data as a 2D array with correct strides. Use the c2c or r2c function with shape_t for dimensions, specify axes like {0,1}, and set the forward parameter. Remember to apply scaling via the fct argument since PocketFFT doesn't normalize results automatically.

PocketFFT vs FFTW: which should I use for my C++ project?

PocketFFT is header-only and C++11 compliant, offering easier integration without linking, while FFTW provides more advanced optimizations and plan varieties. Choose PocketFFT for accuracy and flexibility in pure C++ environments; opt for FFTW if you need peak performance and don't mind external dependencies.

Does PocketFFT support GPU acceleration?

No, PocketFFT is CPU-only and leverages vector instructions for optimization, with no built-in GPU support. For GPU-accelerated FFTs, consider libraries like cuFFT for CUDA or clFFT for OpenCL instead.

How can I enable multi-threading in PocketFFT for better performance?

Multi-threading is enabled by default for multidimensional transforms. Control it using the nthreads parameter in function calls, or disable it by defining the POCKETFFT_NO_MULTITHREADING preprocessor macro as per the configuration options.

What's the memory impact of enabling plan caching in PocketFFT?

Enabling caching via POCKETFFT_CACHE_SIZE can speed up repeated 1D transforms, but the README warns it may use significant memory for large axis lengths. Use it cautiously, especially with very large transforms, to avoid excessive memory consumption.

Can I do in-place real-to-complex FFTs with PocketFFT?

No, in-place transforms are not allowed for r2c and c2r functions. The README specifies that separate input and output arrays are required for these transforms, unlike c2c and r2r which support in-place operation with identical strides.

pocketfft — C++11 Header-Only FFT Library

What is pocketfft?

PocketFFT is a header-only Fast Fourier Transform library for C++11 that provides high-performance, multidimensional FFT, DCT, and DST implementations. It solves the need for a modern, accurate, and flexible FFT library in C++ projects, especially those dealing with scientific computing or signal processing where transform sizes may include large prime factors.

Target Audience

C++ developers working on scientific computing, signal processing, audio analysis, or numerical simulation projects that require efficient Fourier transforms, including support for multidimensional data and various transform types.

Value Proposition

Developers choose PocketFFT for its strict C++11 compliance, excellent numerical accuracy, support for challenging transform sizes via Bluestein's algorithm, and its flexible, header-only design that avoids external dependencies while offering a comprehensive feature set.

Fork of https://gitlab.mpcdf.mpg.de/mtr/pocketfft to simplify external contributions

Use Cases

Best For

Performing multidimensional FFTs on scientific data arrays
Implementing audio or signal processing pipelines in C++
Computing discrete cosine transforms (DCT) for image compression algorithms
Handling FFTs with transform sizes containing large prime factors
Integrating a header-only FFT library into C++11 or later projects
Real-to-complex and complex-to-real transforms with flexible data layouts

Not Ideal For

Applications needing a high-level, simplified FFT API with automatic memory and plan management
Real-time signal processing systems where plan creation latency must be minimized and caching is insufficient
Projects constrained to C++ standards older than C++11 or with strict code size limitations
Developers who prefer extensive documentation, tutorials, and community support over raw performance

Pros & Cons

Pros

Numerical Accuracy Focus

Implements advanced techniques like angle reduction to [0; pi/4] and symmetric evaluations to minimize trigonometric errors, ensuring high precision in transforms as detailed in the README.

Multidimensional Flexibility

Supports transforms on arbitrary axes within N-dimensional arrays with optional multi-threading, making it ideal for scientific data processing where selective axis transformation is needed.

Wide Transform Support

Includes complex FFTs, real FFTs with multiple conventions, DCT, DST types I-IV, and Hartley transforms, covering a comprehensive range of use cases beyond basic FFTs.

Header-Only Integration

As a strictly C++11 compliant header-only library, it's easy to integrate into projects without external dependencies or complex build systems, simplifying deployment.

Cons

Low-Level API Complexity

Requires manual management of strides, shapes, and axes in bytes, which can be error-prone and less intuitive compared to libraries with higher-level abstractions or automatic memory handling.

Limited 1D Threading

Multi-threading is only supported for multidimensional transforms and disabled for 1D transforms, as noted in the README, potentially limiting performance scalability in single-axis applications.

Default Caching Disabled

Plan caching is turned off by default (POCKETFFT_CACHE_SIZE undefined) to save memory, meaning repeated 1D transforms may incur plan creation overhead unless explicitly configured, impacting performance.

Sparse Documentation

The README is technically dense with minimal examples or tutorials, assuming prior FFT knowledge, which can make onboarding difficult for developers new to Fourier transforms.

Frequently Asked Questions

What is pocketfft?

Target Audience

Value Proposition

Use Cases

Best For

Performing multidimensional FFTs on scientific data arrays
Implementing audio or signal processing pipelines in C++
Computing discrete cosine transforms (DCT) for image compression algorithms
Handling FFTs with transform sizes containing large prime factors
Integrating a header-only FFT library into C++11 or later projects
Real-to-complex and complex-to-real transforms with flexible data layouts

Not Ideal For

Applications needing a high-level, simplified FFT API with automatic memory and plan management
Real-time signal processing systems where plan creation latency must be minimized and caching is insufficient
Projects constrained to C++ standards older than C++11 or with strict code size limitations
Developers who prefer extensive documentation, tutorials, and community support over raw performance

Pros & Cons

Pros

Numerical Accuracy Focus

Implements advanced techniques like angle reduction to [0; pi/4] and symmetric evaluations to minimize trigonometric errors, ensuring high precision in transforms as detailed in the README.

Multidimensional Flexibility

Supports transforms on arbitrary axes within N-dimensional arrays with optional multi-threading, making it ideal for scientific data processing where selective axis transformation is needed.

Wide Transform Support

Includes complex FFTs, real FFTs with multiple conventions, DCT, DST types I-IV, and Hartley transforms, covering a comprehensive range of use cases beyond basic FFTs.

Header-Only Integration

As a strictly C++11 compliant header-only library, it's easy to integrate into projects without external dependencies or complex build systems, simplifying deployment.

Cons

Low-Level API Complexity

Requires manual management of strides, shapes, and axes in bytes, which can be error-prone and less intuitive compared to libraries with higher-level abstractions or automatic memory handling.

Limited 1D Threading

Multi-threading is only supported for multidimensional transforms and disabled for 1D transforms, as noted in the README, potentially limiting performance scalability in single-axis applications.

Default Caching Disabled

Sparse Documentation

The README is technically dense with minimal examples or tutorials, assuming prior FFT knowledge, which can make onboarding difficult for developers new to Fourier transforms.

Frequently Asked Questions

pocketfft

What is pocketfft?

Overview

Use Cases

Best For

Not Ideal For

Pros & Cons

Pros

Cons

Frequently Asked Questions

Related Projects

Found a gem we're missing?

pocketfft

What is pocketfft?

Overview

Use Cases

Best For

Not Ideal For

Pros & Cons

Pros

Cons

Frequently Asked Questions

Related Projects

Found a gem we're missing?