Awesome

FFT benchmark

A benchmark for comparison of FFT algorithms performance.

Currently supports

KFR
Intel IPP
Intel MKL
FFTW

CMAKE_PREFIX_PATH should contain the paths to cmake configs of used libraries. Example:

C:/vcpkg/installed/x64-windows-static-md/share
C:/Program Files (x86)/Intel/oneAPI/ipp/2021.9.0/lib/cmake/ipp
C:/Program Files (x86)/Intel/oneAPI/mkl/2024.0/lib/cmake/mkl
kfr-install-dir/lib/cmake

Requires:

Clang 12.0+
CMake 3.12 or newer
AVX2-capable cpu
Python 3.5 or newer
- matplotlib module
- numpy module

Options

Option	Description
`SIZE`	1D FFT
`SIZExSIZE`	2D FFT. Example: `64x32`
`SIZExSIZExSIZE`	3D FFT. Example: `64x32x16`
`--complex flags`	`y` (complex tests), `yn` (all tests), `n` (real tests)
`--inverse flags`	`y` (IDFT tests), `ny` (DFT/IDFT tests), `n` (DFT tests)
`--inplace flags`	`y` (inplace tests), `ny` (all tests), `n` (out-of-place tests)
`--save data.json`	Save results in JSON
`--save -`	Print resulting JSON to stdout
`--avx2-only`	Enable only AVX2 (supported in KFR, IPP, MKL)
`--no-progress`	Disable verbose progress output
`--no-banner`	Disable banner

Benchmark code license

MIT