DADiSP Software Modules - FFT Acceleration
DADiSP/FFTXL
accelerates FFT computations by using the Math Kernel Library from Intel.
Speed improvements by a factor of two are common and prime factor FFTs
(FFTs on series not a power of two in length) experience tremendous speed
gains.
The MKL Library provides highly optimized Fourier transform routines
tuned specifically to Intel processors to provide outstanding performance.
Simply install DADiSP/FFTXL and any routine that employs the FFT or
IFFT function automatically benefits from accelerated computation.
Key Features
- Simple Deployment - just Install and Run
- 2x to 10x FFT speed improvements
- Optimized performance on Intel Processors
- Multi-threaded Execution for even Faster Execution on Multi-Core Systems
- Speeds up any FFT based Analysis
FFT Acceleration Module
The Fourier Transform is a core analysis routine for a wide variety
of technical applications, including signal processing, medical, geophysical,
acoustic, statistics, image processing and many more.
The Discrete Fourier Transform, or DFT, converts discrete time domain
data into discrete frequency domain data. Although discrete FFT Flow
Diagram time data has a continuous Fourier transform, under proper conditions
the DFT effectively samples the continuous transform to enable computer-based
numeric processing of frequency domain data.
The Fast Fourier Transform, or FFT, exploits certain symmetries inherit
to the DFT computation to produce the same result as the DFT but at greater
speeds. The FFTXL module further accelerates DADiSP's built-in FFT function
to deliver one of the fastest FFT computations possible.
Processor Tuned Performance
The FFTXL Module uses the finely tuned Math Kernel Library from Intel
to FFTXL Speed Comparison take advantage of Intel architecture-based
processors using the latest instruction sets, parallelism, and algorithms
to yield a highly optimized FFT function. Performance gains of 2x to
10x over the standard built-in FFT function are acheived.
Simple Deployment
FFTXL is completely automatic, simply install the module and the FFT
function immediately runs faster - no settings to change, no code to
rewrite. In addition to the FFT function itself, any custom or built-in
function that relies on the FFT experiences the same performance gain.
FFTXL is a straightforward way to accelerate any FFT based data analysis.

Requirements
DADiSP/FFTXL requires DADiSP 6.5 B01 or higher. Contact us for information
about updating your current version of DADiSP.
FFTXL FFT Acceleration Module
FFTXL automatically accelerates the computation of the FFT and IFFT
functions. In addition, custom or built-in routines that make use
of the FFT experience significant speed improvements. Below is a
partial list of routines provided by DADiSP that benefit directly
from the FFTXL Module. In addition, many of the functions provided by the
DADiSP/Filters and DADiSP/AdvDSP modules also benefit
from the speed optimizations offered by FFTXL.
| FFT Based Functions |
| cceps |
Complex cepstrum |
| dct |
Discrete cosine transform |
| dct2 |
2D Discrete cosine transform |
| decilp |
Bandlimited decimation |
| facorr |
Auto-correlation function |
| facov |
Auto-covariance function |
| fcirconv |
Fast circular convolution |
| fconv |
Fast convolution |
| fdeconv |
Fast deconvolution |
| fft |
Fast Fourier Transform |
| fft2 |
2D Fast Fourier Transform |
| finteg |
Frequency domain integration |
| firsamp |
Frequency sampling filter design |
| fpadfilt |
Filtering with end padding |
| fxcorr |
Cross-correlation function |
| fxcov |
Cross-covariance function |
| fzinterp |
Interpolation by FFT zero insertion |
| grpdelay |
Group delay |
| hilb |
Hilbert transform |
| idct |
Inverse discrete cosine transform |
| idct2 |
2D Inverse discrete cosine transform |
| ifft |
Inverse Fast Fourier Transform |
| ifft2 |
2D Inverse Fast Fourier Transform |
| invpsd |
PSD to time series |
| nfft |
Zero pad or time aliased FFT |
| npsd |
Zero pad or time aliased PSD |
| nspectrum |
Zero pad or time aliased spectrum |
| powspec |
Power spectrum |
| psd |
Power spectral density |
| specgram |
Joint time-frequency spectum |
| sfreq |
S transform evaluation |
| spectrum |
Normalized FFT magnitude |
| rceps |
Real cepstrum |
| zfreq |
Z transform evaluation |
|