A basic C# wrapper for FFTS (The Fastest Fourier Transform in the South).

FFTSSharp is MIT licensed, so personal and commmercial use alike is allowed. For further details, see the LICENSE file.

The dynamically linked libraries (DLLs) used for the wrapper are custom built with CMake and should work normally for Windows users. The DLLs are included in the package and the folder "ffts-dlls" is to be put at the application's startup path.

Features

• Full support over all 3 types of SSE (SSE1, SSE2, SSE3) and even without SSE (nonSSE). A built-in SSE availability detector takes care of this.
• Full managed function calls to FFTS unmanaged functions
• Easy-to-use wrapper around FFTS with high performance
• Test programs for basic usage and benchmark code

Benchmark

This is a benchmark comparing FFTSSharp against Intel's MKL C# implementation (https://software.intel.com/content/www/us/en/develop/articles/using-intel-mkl-in-your-c-program.html).

Tests on sizes of power of 2 with complex-to-complex transforms. Transforms for each sizes are run 100000 times and the results are averaged out.

The benchmark is carried out on i7-2640M CPU; SSE3 for FFTSSharp.

Length	FFTS	MKL
256	356 ns	376 ns
512	1551 ns	1555 ns
1024	1551ns	1555 ns
2048	3949 ns	12520 ns
4096	8342 ns	22965 ns
8192	20423 ns	45957 ns
16384	46817 ns	89313 ns
32768	90406 ns	188335 ns

Target

.NET Framework 4.5 (or .NET Standard 2.0) and above
For target of .NET Framework 2.0 - 3.5 please refer to FFTSSharpLegacy

Get FFTSSharp

For ease of installation, you can add the lastest release of FFTSSharp into your project using Nuget

Usage

• Include the library

using PetterPet.FFTSSharp;

• Initial loading of FFTS DLLs

FFTSManager.LoadAppropriateDll(FFTSManager.InstructionType.Auto);

There consists of 4 enums in FFTSManager.InstructionType: nonSSE, SSE, SSE2, SSE3, Auto
(Choose "Auto" for generic loading)

• Create an instance of FFTS (think of it like creating a plan)

var ffts = FFTS.Real(FFTS.Forward, 16);

• Declare the input and output array (length of the input and output must be known beforehand)

float[] input = new float[16];
float[] output = new float[18];

• Compute the result

ffts.Execute(input, output);

• Dispose the plan after use

ffts.Dispose();

Current problems

• For now, only 1D real and complex, forward and inverse tranforms are supported. FFTS support for 2D and above is insufficiently documented and is incomplete.

Credits

Much thanks to the author, anthonix for creating FFTS with a very permissive license.
Thanks to DllImportX made available by rodrigo-speller.

Thoughts

Contributions and issue reports are welcomed.