SleefDFT faster than FFTW?

[shibatch]

Below is the result of performance comparison between SleefDFT and FFTW3.
As shown, SleefDFT now looks like faster than FFTW.

The vertical axis indicates Mflops calculated in the way indicated in the FFTW web site.
The horizontal axis indicates log2 of the size of transform.

Compiler : gcc version 14.2.0 (Ubuntu 14.2.0-4ubuntu2~24.04)
CPU : Ryzen 9 7950X (clock frequency fixed at 4.5GHz)
Build option : -DSLEEF_BUILD_DFT=True -DSLEEFDFT_ENABLE_STREAM=True -DSLEEFDFT_MAXBUTWIDTH=7
FFTW is the one on Ubuntu 24.04.2.

Execution plans were made with SLEEF_MODE_MEASURE mode and FFTW_MEASURE mode, respectively.
Measurements were taken in the single thread mode.

I hope I am not mistaken in the measurements.

[shibatch]

Below is the data with m1 macbook pro with Asahi linux, ubuntu.

The settings are almost the same. The data type of the transform is double precision.

[shibatch]

Core i5-11400, ubuntu

[shibatch]

Here is a result with multithreaded runs on Ryzen 9 7950X.
Not too bad. I can now say that SleefDFT is comparable to FFTW.

[shibatch]

Core i5-11400, ubuntu, double precision

Core i5-11400, ubuntu, single precision