Compare KFR with other libraries

KFR is a C++20 DSP framework for vectorized math, FFTs, filters, resampling, and tensor operations. The comparison below puts its FFT and DSP interface beside popular alternatives.

Feature coverage follows the supported-libraries table in the FFT benchmark README. These are capability comparisons, not a promise that one library is fastest for every workload. Measure your own signal sizes, data types, CPU, and memory layout.

View KFR benchmark results Open kfrlib/fft-benchmark

Feature comparison

FFT and sample-rate conversion feature comparison
Library Version Float Double 1D 2D & 3D Arbitrary sizes Notes Runtime dispatch
KFR n/a Yes Yes Yes Yes Yes Yes
FFTW 3.3.10 Yes Yes Yes Yes Yes Yes
Intel MKL n/a Yes Yes Yes Yes Yes Yes
Intel IPP n/a Yes Yes Yes No Yes Yes
Sleef 3.9.0 Yes Yes Yes No No Power-of-2 sizes only Yes
PFFFT 1.0.0 Yes No Yes No No Multiples of SIMD size2 No
JUCE 8.0.7 Yes No Yes No No Yes
KissFFT 131.2.0 Yes Yes Yes No Yes No real inverse; out-of-place only No
OTFFT 11.5.0 No Yes Yes No Yes x86 only, real requires even sizes No
PocketFFT 2023-09-25 Yes Yes Yes No Yes No
vDSP (Accelerate) n/a Yes Yes Yes No Yes Apple only n/a
libsamplerate 0.2.2 Yes No n/a n/a n/a SRC only No
soxr 0.1.3 Yes Yes n/a n/a n/a SRC only No
speexdsp 1.2.1 Yes No n/a n/a n/a SRC only No
r8brain-free-src 7.1 Yes Yes n/a n/a n/a SRC only No

Results and support depend on the wrapper and configuration used by the benchmark. For example, the benchmark notes restrictions such as power-of-two sizes, SIMD-size multiples, Apple-only availability, or missing real inverse transforms.

For measured FFT results, test conditions, and the complete library list, see the KFR benchmark page and the kfrlib/fft-benchmark repository.