kissfft/README
2003-08-09 00:59:17 +00:00

80 lines
2.9 KiB
Plaintext

KISS FFT - A power-of-two Fast Fourier Transform based up on the principle,
"Keep It Simple, Stupid."
There are many great fft libraries already around. Kiss FFT is not trying
to be better than any of them. It only attempts to be a reasonably efficient,
moderately useful FFT that can use fixed or floating data types and can be
incorporated into someone's C program in a few minutes with trivial licensing.
USAGE:
The basic usage is:
void * cfg = kiss_fft_alloc( nfft ,inverse_fft );
while ...
...
kiss_fft( cfg , cx_buf_in_out );
...
free(cfg);
Declarations are in "kiss_fft.h", along with a brief description of the
two functions you'll need to use. Code definitions are in kiss_fft.c, along
with sample usage code.
The code can be easily recompiled to work with 16bit fixed point data,
or various floating point types. The default is float.
BACKGROUND:
I started coding this because I couldn't find a fixed point FFT that didn't
use assembly code. I started with floating point numbers so I could get the
theory straight before working on fixed point issues. In the end, I had a
little bit of code that could be recompiled easily to do ffts with short, float,
or double (other types should be easy too).
Once I got my FFT working, I wanted to get some performance numbers against
a well respected and highly optimized fft library. I don't want to criticize
this great library, so let's call it FFT_BRANDX.
During this process, I learned:
1. FFT_BRANDX has 500 times as many lines of code as Kiss
(and that's just the C code).
2. It took me an embarrassingly long time to get FFT_BRANDX working.
3. FFT_BRANDX is almost 3 times faster than Kiss
It is wonderful that free, highly optimized libraries like FFT_BRANDX exist.
But such libraries carry a huge burden of complexity necessary to extract every
last bit of performance.
Sometimes simpler is better, even if it's not better.
PERFORMANCE:
(on Athlon XP 2100+, with gcc 2.96, optimization O3, float data type)
Kiss performed 1000 1024-pt ffts in 136 ms. This translates to 7.5 Msamples/s.
Just for comparison, it took md5sum 160 ms to process the same amount of data
DO NOT:
... use Kiss if you need the absolute fastest fft in the world
... use Kiss if you need mixed radix FFTs
... ask me to add features that will bloat the code
UNDER THE HOOD:
Kiss uses a complex-only, frequency decimation, radix 2, in-place FFT. Bit reversed
addressing is corrected as the last step in the transform. No scaling is done.
LICENSE:
BSD, see COPYING for details.
TODO:
Make the fixed point scaling and bit shifts more easily configurable.
Document/revisit the input/output fft scaling
See if the fixed point code can be optimized a little without adding complexity.
AUTHOR:
Mark Borgerding
mark@borgerding.net