Autumn sale: up to 25% off (22 OCT - 1 NOV)

FFT benchmarks

Performance measurements for KFR's forward FFT implementations on x86-64 AVX2 and ARM NEON.

The benchmark tool and test harness are available in kfrlib/fft-benchmark.

Compare KFR with other libraries

Contents

AVX2 · NEON

AVX2, Intel x86-64

Float real forward FFT

In-place
AVX2 float real forward in-place FFT benchmark
Out-of-place
AVX2 float real forward out-of-place FFT benchmark

Float complex forward FFT

In-place
AVX2 float complex forward in-place FFT benchmark
Out-of-place
AVX2 float complex forward out-of-place FFT benchmark

Double real forward FFT

In-place
AVX2 double real forward in-place FFT benchmark
Out-of-place
AVX2 double real forward out-of-place FFT benchmark

Double complex forward FFT

In-place
AVX2 double complex forward in-place FFT benchmark
Out-of-place
AVX2 double complex forward out-of-place FFT benchmark

NEON, Apple Silicon

Float real forward FFT

In-place
NEON float real forward in-place FFT benchmark
Out-of-place
NEON float real forward out-of-place FFT benchmark

Float complex forward FFT

In-place
NEON float complex forward in-place FFT benchmark
Out-of-place
NEON float complex forward out-of-place FFT benchmark

Double real forward FFT

In-place
NEON double real forward in-place FFT benchmark
Out-of-place
NEON double real forward out-of-place FFT benchmark

Double complex forward FFT

In-place
NEON double complex forward in-place FFT benchmark
Out-of-place
NEON double complex forward out-of-place FFT benchmark

These are preliminary results from our benchmark suite. We are gathering data from various machines. Full benchmark results, including raw data, will be available soon.